Image Processing, Video Terminals, and Printer Techuologies




Digital Technical Journal
                       Digital Equipment Corporation




                                            Volume 3 Number 4

                                                     Fall 1991
                                                                  Editorial
                                                                  Jane C. Blake, Editor
                                                                  Helen L. Patterson, Associate Editor
                                                                  Kathleen M. Stetson, Associate Editor
                                                                  Leon Descoteaux, Associate Editor

                                                                  Circulation
                                                                  Catherine M. Phillips, Administrator
                                                                  Sherry L. Gonzalez

                                                                  Production
                                                                  Mildred R. Rosenzweig, Production Editor
                                                                  Margaret L. Burdine, Typographer
                                                                  Peter R. Woodbury, Illustrator

                                                                  Advisory Board
                                                                  Samuel H. Fuller, Chairman
                                                                  Richard W Beane
                                                                  Robert M. Glorioso
                                                                  Richard]. Hollingsworth
                                                                  John W McCredie
                                                                  Alan G. Nemeth
                                                                  Mahendra R. Patel
                                                                  F Grant Saviers
                                                                  Victor A. Vyssotsky
                                                                  Gayn B. W inters

                                                                  The Digital Technicaljournal is pub Iished quarterly by Digital
                                                                  Equipment Corporation, 146 Main Street ML01-3/B68, Maynard,
                                                                  Massachusetts 01754-2571. Subscriptions to thejournal are $40.00
                                                                  for four issues and must be prepaid in U.S. funds. University and
                                                                  college professors and Ph.D. students in the electrical engineering
                                                                  and computer science fields receive complimentary subscriptions
                                                                  upon request. Orders, inquiries, and address changes should be
                                                                  sent to the Digital Technicaljournal at the published-by address.
                                                                  Inquiries can also be sent electronically to DTJ@CRLDECCOM.
                                                                  Single copies and back issues are available for $16.00 each from
                                                                  Digital Press of Digital Equipment Corporation, 1 Burlington
                                                                  Woods Drive, Burlington, MA 0 1 80 3-45 39.

                                                                  Digital employees may send subscription orders on the ENET to
                                                                  RDVAX::)OURNAL or by interoffice mail to mailstop MLOI-3/B68.
                                                                  Orders should include badge number, site location code, and
                                                                  address. All employees must advise of changes of address.

                                                                  Comments on the content of any paper are welcomed and may
                                                                  be sent to the editor at the published-by or network address.

                                                                  Copyright© 1991 Digital Equipment Corporation. Copying
                                                                  without fee is permitted provided that such copies are made for
                                                                  use in educational institutions by faculty members and are not
                                                                  distributed for commercial advantage. Abstracting with credit
                                                                  of Digital Equipment Corporation's authorship is permitted.
                                                                  All rights reserved.

                                                                  T he information in thejournal is subject to change without
                                                                  notice and should nor be construed as a commitment by Digital
                                                                  Equipment Corporation. Digital Equipmem Corporation assumes
                                                                  no responsibility for any errors that may appear in thejournal.

                                                                  lSSN 0898-901X

                                                                  Documentation Number EY-H889E-DP

                                                                  The following are trademarks of Digital Equipment Corporation:
                                                                  ALL-IN-1, DECimage, DECnet, DECprint, DECserver, DECstarion,
                                                                  DECwindows, Digital, the Digital logo, LAT, LN03, MicroVAX,
                                                                  PrintServer, Q-bus, ReGIS, rtVAX, ULTRIX, VA,'\, VAXELN,
                                                                  VAXstation, VMS, VT1000, VTI200, VT1300, and VXT 2000.

                                                                  Apple DeskTop Bus is a trademark and LocalTalk is a registered
                                                                  trademark of Apple Com purer, Inc.

                                                                  Motorola and 68000 are registered trademarks of Motorola, Inc.

                                                                  Open Software Foundation is a trademark and OSF and OSF/1 are

Cover Design                                                      registered trademarks of Open Software Foundation, Inc.

High-performance screen display of bitonal images is one          PostScript is a registered trademark of Adobe Systems, Inc.
of the topics in this issue. The handwriting and manually         Texas Instruments is a trademark of Texas Instruments, Inc.
produced technical drawings on our cover are types of images
                                                                  UNIX is a registered trademark of UNIX System Laboratories, Inc.
that can be scanned, st01·ed electronically, ana then displayed
                                                                  X W indow System is a trademar.k of the Massachusetts Institute
on an X terminal screen; portions of an image can be enlarged
                                                                  of Technology.
or rotated on screen.
                                                                  Book production was done by Digital's Database Publishing Group
The cover was designed try Sandra Calef of Calef Associates.      in Northboro, MA.
I Contents
 7    Foreword
      Larry Cabrinety


                                    Image Processing, Video Terminals, and Printer Technologies




  9   Hardware Accelerators for Bitonal Imag e Pro cessing
      Christopher ). Payson, Christopher ]. C ianciolo,
      Robert N. Crouse, and Catherine F. Winsor

 26   X Window Terminals
      Bjorn Engberg and Thomas Porcher

 36   ACCESS. bus, an Open Desktop Bus
      Peter A. Sichel

 43   Design of the DECprint Comm on Printer Supervisor
      for VMS Systems
      Richard Landau and Alan Guenther

 55   T he Comm on Printer Access Protocol
      James D. Jones, Ajay P Kachrani, and Thomas E. Powers

 61   Design of the Turbo PrintServer 20 Controller
      Guido Simone, Jeffrey A. Metzger, and Gary Vaillette




                                                                                                  1
              I   Editor's Introduction

                                                          boards, mice, and t ablets, a l l of which m ay not be
                                                          made by the same company. A new open desktop
                                                          bus, described by Peter Sichel, is a simple means to
                                                          connect as many as 14 l ow-speed devices to a desk­
                                                          top system. In his pap er, Peter presents the project
                                                          background, reviews the I 'C technology on which
                                                         the bus is based, and describes the protocol and the
                                                          configuration process.
                                                            Hard-copy presentation of data and recent deve l ­
                                                         opments i n printer tec hnologies are the topics of t he
                                                         next th ree papers. Rick Landau and Alan Guenther
                                                         review t he DECprint Printing Services, whi c h is
                  jane C. Blake
                                                         software that controls nu merous printer features
                  Editor
                                                         for a wide range of printers. Also cal led a common
                                                         print symbiont, this component of t he VMS print­
 Products designed for quali ty, high-performance        ing system supports several p age description lan­
presentation of data in both video and hard-copy         guages, hand les m ultiple media simul taneously,
form are the topics of p apers in this issue of the      and uses different I /0 interconnections and com­
Digital Technical journal. The design cha l lenges       n mnication protocols.
range from managing the h uge storage require­              Both DECp rint Printing Services and the s ubject
ments of images for display on X terminals to ensur­     of the next pap er, the common printer access pro­
ing h igh-perform ance in a feature-rich printer         tocol, are part of the DECprint architecture. The
environment.                                             CPAP provides t he fund amental services necessary
    Image processing is the subject of t he opening      for the presentation of data at the printer. Jim
paper by Chris Payson, Chris C ianciolo, Bob Crouse,     Jones, Ajay Kachrani, and Tom Powers describe the
and Cathy Winsor. The a uthors note that one advan­      c hallenge s of developing a protocol that operates
tage of scanning images for screen disp l ay is t he     in a heterogeneous, internetworking environment
i n p ut time saved; however, t he scanned i mages       and that also ensures backward compa tibil ity with
and data can consume sign i ficant amou nts of stor­     older pro d ucts. Their success in developing a high­
age space. They then review the development of an        performance protocol is evidenced by OSF accep­
image accelerator board that not only helps solve        tance of CPAP for inclusion in a future release of
the problem of storage but also addresses t he need      OSF/ 1 .
for high-performance display-view and manipu la­            A s was the case with the CPAP, perform ance
tion-of bitonal i mages. In addition to specifics of     was also key in the development of the t urbo
the board implementation, the authors offer an           PrintServer 20 controller. Guido Simone , Jeff Metzger,
overview of imaging concepts, terms, and future          and Gary Vai l lette explain that the req uirements of
directions for image accelerators.                       complex docu ments demanded turbo controller
    T he terminal on which the image accelerator         performance that was five to eight times that of the
board resides is D ECimage 1200, an X terminal.          c urrent control ler. To a i el t hem in making design
X terminals development in general , incl uding a        decisions, a performance analysis tool, RETrACE,
discussion of the VT1200, is the sub ject of a paper     was created and is descri bed here . Authors also
by B j orn Engberg and Tom Porcher. Bjorn and Tom        relate how they used existing chips in order to keep
focus their d iscussion on a comparison of the           development costs low and still del iver a high­
X terminal and X workstation environments, and           p e rformance control ler.
explain why X termi nals are a low-cost alterna tive .      The editors thank Liz Griego-Powel l of the Video,
The au thors present the design choices debated by       I m age and Print Systems Gro up for her help i n
the engineers during the development of Digital's        preparing this issue.
X terminals , including t he selection of a hardware
plat form, terminal and window management,
X server, com mu nications protocols, and fo nt fi le
systems.
    Video terminal and workstation users need the
assistance of a n umber of 1/0 devices, such as key-



2
                                                             Biographies                    I

Christopher J. Cianciolo      As a hardware design engineer in the Video, Image
and Print Systems Gro up , Chris C i anciolo is currently working on the design
for the group's l atest imaging product. Chris jo ined D igital in 1985 after par­
ticipating in a co-op session in the Power Supply Engineering Gro up . He also
participated in co-op sessions for Charles Stark Draper Laboratory, Inc. on a
fiber-optic m issile guidance system project. He received his B . S.E.E. from
Northeastern University in 1988 and is c urrently pursui ng an M . S.E.E. , also from
Northeastern.



Robert N. Crouse       Senior engineer Bob Crouse is a member of the Video ,
Image and Print Systems Group. He is currently working on the advanced devel­
opment of new imaging technology for X window terminals. Bob was project
engineer for the development of a bitonal imaging accelerator for a low-end
VAXstation workstation. As a member of the Electronic Storage Development
Group , he designed a double-bit error detection and correct ion circuit for a VAX
mainframe. Bob received his B.S.E.E. from Northeastern University and h o lds one
patent.



Bjorn Engberg As a principal software e n gineer in the Video, Image and Print
Systems Group, Bj orn Engberg was the main architect and software project
leader for the VTIOOO and VT1200 X window term inals. He joined Digital in 1978
and worked as a development engineer a t CSS i n Sweden, where he modified
D igital's terminals for the European m arket. He relocated to the United States
in 1982 to work on the VT240, the VT320, the LJ250, and several advanced devel­
opment projects. Bj orn received an M . S.E.E. (honors) from the Royal Institu te of
Technology in Stockholm.



                                                                                        3
Biographies




              A. Alan Guenther          As a member of the technical st aff i n the DECprint System
              Software Group, Al an Guenther is involved in the ongoing design and implemen·
              t at ion of the DECprint common print symbiont. Prior to this, he was the prim ary
              designer and implementor of the d istri buted queuing services. Alan h as worked
              at Digi t al since 1973, both as a fu ll-time employee and as an independent consul­
              t ant (from 1982 to 1990). After receiving a B.S. 01onors, 1970) from the University
              of Mo man a, he worked at the university until he joined D igi tal.




              James D. jones       Jim Jones is a princip al engineer in the H ardcopy Systems
              Engineering Grou p . He joined D igital in 1974 and was p ar t of a team developing
              diagnostic programs for the DECsystem-10 and DECSYSTEM-20 systems. After run­
              ning his own software business for five years, Jim rejoined D igi t al to design
              printer contro llers and software. Most recently, he provided software for t h e
              PrintServer products, authored the Common Printer Access Protocol speci fica­
              tion, and is helping to define the next generation of network printers. Jim is a
              member of IEEE and ACM and p articip ates i n the JETF.



              Ajay P. Kachrani          Princip al software engineer Ajay Kachrani cu rrently works
              on the OSF/1 socket and XTI kernel interfaces and security proj ect. Previously ,
              he led the development of the overal l PrintServer software version 4.0 with
              d u al network protocol support (DECnet and TCP/IP), from inception through
              field test. Aj ay presented the CPAP protocol as an Internet st and ard to the IETF
              and added PrintServer support in version 1 .0 of the P al l adium Print System at
              MIT/Project Athen a. Aj ay holds a B.S.E.E. (honors) from the University of Mysore,
              Ind i a, and an M .S.C.S. from the University of Lowell.



              Richard B. Landau Richard L an d au is the DECprint program m an ager for the
              Video, I m age and Print Systems Group . Working to i mprove the interaction of
              printing soft w are and hardware, he initiated the DECprint, Font, and PostScrip t
              programs. Prior to this, Rick w as the program and development m an ager for the
              VAX DBMS, DATATRlEVE, COD, and Rdb/VMS products and for the rel ational
              d at ab ase architecture. Before joining Digit al in 1974, Rick w as an independent
              consu l t ant and was also employed by Appl ied D at a Research, Inc. He holds A.B.
              (cum l aude, 1969) and M.A. (1973) degrees in statistics from Princeton University .



              Jeffrey A. Metzger     Presently a senior engineer, Jeff c ame to Digi t al as a co-op
              student in 1983, working first in the Semiconductor Engineering Group and then
              in H ardcopy Engi neering. He bec ame a fu ll-time employee after gradu ating from
              Cornel l University in 1985. He introduced H ardcopy to system-level logic simu­
              l ation, contributed to the hardware, software, and firmware development of the
              PrintServer 20, an d developed R ETrACE, which is used to ch aracterize the exe­
              cution behav ior of PrintServer systems. Jeff is currently working in the Entry
              Systems Busi ness Group on a n e xt-generation processor product.


4
                                                                                           I
Christopher J. Payson         Chris Payson joined D igital as a hardware design
engineer in 1989 after five co-op terms. He is currently working on XIE software
and i m age hardware accelerators. Chris previously worked on performance
testing, d iagnostics, logic design, and demonstration software, all associatt:d
with imaging. He is coapplicant for a patent related to an image clipping algo­
rithm and hardware logic. Chris received a B.S.C.E. from Rochester Institu te of
Technology with highest honors and is currently pursu ing an M.S.C.E. from
Northeastern University.



Thomas C. Porcher       Principal engineer Tom Porcher is a member of the
Video, Image and Print Systems Group. He provided technical leadership in the
development of the VXT 2000 X terminal. Previously he was a technical leader
for the VT240 terminal, VAX Session Support Utility, and the DEctt:rm terminal
emulator. Tom holds five patents for work on the VT240 terminal and on the multi­
session protocol used in the VT340 and VT400 series terminals. Tom received his
B.S. in mathematics from Stevens Institute of Technology (1975). Ht: is a member
of the ACM.



Thomas E. Powers      As a consultant engineer in the Hardcopy Engineering
Firmware/Software Group, Tom Powers is a vendor l iaison for desktop
PostScript printer products. He chairs the DECprint PAP Architecture Team
and was a contributor to the PrintServer 40 internal hardware/firmware archi­
tecture. Tom represented Digital on American and intern ational standards com­
mittees on computer graphics from 1979 to 1989. He led several firmware teams
and is coinventor of the ReGIS Graphics Protocol. Tom has a B.S. E. E. from Tufts
University and an M .S.E.C.E. from the University of Massachusetts at Amherst.



Peter A. Sichel  As a principal software engineer in the Video Terminals Archi­
tecture Group, Peter Sichel led the development of the ACCESS. bus architecture
and device protocol specifications, in addition to writing the init i al ACCESS. bus
device firmware. He worked on the VT420 video terminal and the DECterm
DECwindows terminal emulator, and helps maintain D igital standards for video
terminals and keyboards. Peter joined Digital in 1981 after receiving B.S. and M .S.
degrees in computer engineering from the University of Michigan.




Guido R. Simone     G u ido Simone is a principal engineer in the Print Systems
Engineering Group and was the project leader and architect for the turbo
PrintServer 20 controller. He is currently working on the development of a
new print system architecture to be used with advanced printing technologies.
In previous work, G u ido was the project leader and architect for an rtVAX
78R32 CPU chip-based laser printer controller. Before joining Digital in 1980, he
received a B.S. in electrical engineering from Rensselaer Polytechn ic Institute.




                                                                                       5
Biographies




              Gary P. Vail lette   Senior hardware engineer Gary Vai l lette has heen involved
              in the design and implementation of printing system hardware since joining
              Digi tal in 1983. His current work incl udes performance characterization of
              PostScript printers and PrintServer products, and hardware implementation of
              cern decomp ress ion in the t u rbo PrintScrver 20 product. Previously, Gary
              worked at Data General Corporation and helped to develop their token bus
              network product. He holds an A.A.E.E (1974) from Quinsigamond Com m u n i ty
              College and expects to receive a B.S.C.S. (May 1992) from Boston Universi ty.



              Catherine F. Winsor      As a senior engineer in the Video, Image and Print Sys­
              tems Group, Cathy Winsor has worked on image accelerators. As the project
              leader for the DEC:image 1200 hardware and the image u t il i ty l ibrary software,
              Cathy was involved i n the planning and development of an i mage-capable
              VT1200. She is curre n t ly leading the project to support im aging on the next
              generation of Digital's X terminals. The project i nclucles an image accelerator
              board and XIE software. Cathy received an A.B. i n engineering sciences from
              Dartmouth Co l lege and a B.S.E.E. from the Thayer School of Engineering.




6
              I   Foreword

                                                           processing. The X terminal user can now benefit
                                                           from the graphical user interface, sophisticated
                                                           appl ications, and standards of performance previ­
                                                           ously available only on workstati ons. X terminals
                                                           run X l l server code which is operating system
                                                           independent and ideal ly suited for heterogeneous,
                                                           network-based computing environments. In this
                                                           issue you wil l read about the engineering decisions
                                                           made as the X terminals were developed.
                                                              There is a growing need in the industry to have
                                                           imaging applications run alongside conventional
                                                           text and graphics appl ications. Technical docu­
                  Larry Cabrinety
                                                           mentation is an example of this. Imaging appl ica­
                  Vice President,
                                                           tions, however, have special requirements to achieve
                  Video, Image and Print
                                                           acceptable end-user performance. Although the X l l
                  Systems Group
                                                           software can handle images a s bit-map data, soft­
                                                           ware and hardware assistance is requ ired to achieve
                                                           acceptable p erformance. Digital has designed
For the m i l l ions of people worldwide who use           DECimage hardware accelerators for rapid process­
Digital's computer equipment, the computer is not          ing of i mage data. This technology is included in
the sophisticated system in the back room, or the          the DECimage 1200 and will be incorporated in
complex network. It is the equ ipment they use             fol lowing generations of X terminals. To make this
each day- the terminal or monitor, keyboard and            possible, Digital developed extensions to the
mouse, desktop printer or network printer system.          X server software that support the high-speed
   Today's users demand products with h igh levels         transport and display of image data. To assure open
of usabil ity and superior ergonomic features.             standards, the extensions have been proposed to
D igital's products set worldwide standards for the        M IT for incorporation into releases of the X l l
user i nterface to computer systems. In the 90s our        server software.
focus is to offer products that operate i n m u lti­           In November 1990, D igital announced its next
vendor environments with the goal of delivering a          generation of X terminals. The VXT2000 terminal
complete compu ting sol ution. In this issue you will      provides virtual memory and supports both a tra­
read about some of the Video, Image and Print Sys­          ditional host-based model with software down­
tems (VIPS) Group's products and technologies that         loaded to the terminals as well as the server style of
support network computing and standards-based              X terminal compu ting.
environments.                                                  The VXT2000 terminal was designed to support
   D igital entered the video terminal market in 1975      TCP/IP and LAT protocols, and further demonstrates
with the VT52 for i t s time-sharing users. Its replace­   our comm itment to openness and support for cus­
ment, the VT100, embodied two important princi­             tomers' m u lt ivendor environments. Th is same phi­
ples-the use of standards in data com munications           losophy is seen in our printer products and our open
in terchange and the protection of customer invest­         desktop bus.
ments through backward compatibility of new gen­               Digi tal pioneered the distri bu ted printing busi­
erations of products. The VT220, introduced in              ness with networked laser printers. Th is prod­
1983, and the cost-effective VT320 terminals saw           uct area began when we combined two concepts
the addition of functional i ty and ergonomic fea­          which had not been combined before-mid- range
tures which establ ished D igi tal as a leader i n the     l aser printers and networks. I n the mid-1980s most
commodities market.                                        large-scale comp uting was done on mainframe
   I n March 1990, Digital entered the X terminal          compu ters with l arge printers attached directly
market with the i ntroduction of the VTlOOO, fol­          to these systems. Typically these dedicated print­
lowed by the VT1200 and VT1300 terminals later              ers were only accessible to users on that particular
that year. The emergence of MIT's X Window                  system. D igital's distributed compu ting provided
Systems as the accepted industry standard for               an alternative to the ma inframe. By combi ning the
windowing systems provided a standards-based               power of multiple systems in c lusters or on net­
environment for distri buted applica tions display          works, a new distribu ted large system was created.



                                                                                                               7
Foreword




A printing sol ution was needed to effectively work           original PrintServer 40, a l ong with cu stom hard·
in t h i s new d i stribu ted computing environment.          ware acceleration hoards developed by the liard­
The Print Server serie s add n:ssed this n eed.               copy Group to enable printing a t 40 pa gt:s per
    PrintServer products enabled printing resources            mi nute. In this issue you wi l l read about t h <.: single·
 to be directly connected to networks for the fi rst          board control ler that replace s the Mic roVAX 11 and
 time, and since they were on the network and not             offe r s far more processing power. Using the latest
tied to any one system, they were acce ssible hy a l l        system-on-a-chip technology, our new turbo board
systems o n those n etworks. They enabled the com·            provides lea d ership p erformance for our prin ters.
plex printer fu nctiona l i ty previously found only in       The CCITI image decompression chip enables us to
dedicated mainframe printers to be d i st ribu ted            provide fu ll-speed image printing to our customers
throughout end-u ser environments.                            as the image market develops.
   As these mid-range printers migrated out of the               The first PrintServer systems supported printing
computer room and into the office, new demands                from VMS hosts ovt:r D ECnet networks. Since then
for funct iona lity were created. Large groups of users       the breadth of platform support has i ncreased to
brought many d ifferent requi rement s for printing,          include first ULTRJX system s and then UNIX operat·
and our goal was to sat isfy as m any as possi ble in a       ing systems. A software kit for Sun systems wil l be
single PrintServer. For example , some people need            available soon. In expanding PrintServer connectiv·
"A" size paper for office c orrespondence, while              ity to include UNIX systems and TCP/11' network s,
others may n eed "B" size paper for CAD/CAl\1 or              we again faced the problem that no network print·
accoun ting work, and stil l others need transparen­          ing protocol existed for TCP/IP. With the h elp of
cies for presentation s. The PrintServer is flex ible         D igital's experts at the We stern Re search Labora·
enough to have a l l of these d ifferent types of             tory, we were able to devdop a solution. In this
media available and offer both simplex and duplex             issue, we discuss the creation of a network printer
printing.                                                     access protocol for TCP/ IP. Today this network pro­
   In 1985 when Digital was first developing the              tocol is a proposed standard at the Internet Engi·
PrintServer, there was no industry standard way               neering Task Force, the body control l ing the TCP/IP
of describing the conten ts of a page to a print er.          protocol.
Each m ajor vendor had its proprietary language,                 The development of the ACCESS. bus product has
and none offered the compa tibil ity necessary to             brought an easy, st andard way to l i nk a desktop
achieve our print system vision. Our goal was to              computer to many interactive u ser interfaces. This
create a fam ily of produc t s, from large to small,          open desktop bus is currently implemented on the
that offered compa t i bi l i ty for all app l ications. To   Personal DECstation 5000 wor k station, and im ple·
achieve this goal we had to select a protocol                 mentations on future RISC workstations and video
that would enable us to print any file on any                 terminals is u nderway. Developers of Digita l's prod·
printer. At that time Adobe Systems, the developer            u cts wi l l continue to pl ace a h igh priority on open
of PostScript, was a s m a l l start-up company in            standards. The papers included i n this issu t: of the
Silicon Val ley . PostScript was not a standard, and in       Digital Technicaljournal will providt: insight into
fact, only a single PostScript laser printer model            the key areas of technology u st:d in the design and
had been shipped, the original Apple LaserWriter.             development of VJPS product s.
Our technical co mmunity fe lt PostScript was the
best s olution to ou r net:ds, and at that point Digi tal
com mitted to adopting PostScript as our strategic
page description language. PostScript printers and
PostScript application support are now pervasive
throughout the industry and standard printing
protocols enable interactive com mun ication with
hosts on the network.
   Signi ficant advances have taken place in the
PrintServer series over the past seven years. An
entire :VlicroVAX II system was housed wi thin the



8
                                                                                       Christopher]. Payson
                                                                                    Christopher]. Cianciolo
                                                                                             Robert N. Crouse
                                                                                          Catherine E Winsor




Hardware Accelerators for
Bitonal Image Processing

               Electronic imaging systems transfer views of real-world scenes or objects into
                digital bits for storage, manipulation, and viewing. In the area of bitonal images,
                a large market exists in document management, which consists of scanning vol­
                umes of papers for storage and retrieval. However, high scan densities produce
               huge volumes of data, requiring compression and decompression techniques to pre­
                serve system memory and improve system througbput. These tecbniques, as well as
                general image processing algorithms, are compute-intensive and require high
                memory bandwidth. To address tbe memory issues, and to acbieve interactive
               image display petformance, Digital bas designed a series of bitonal image hard­
                ware accelerators. Tbe intent was to create interactive media view stations, with
               imaging applications alongside other applications. In addition to achieuirzg mem­
                ory, performance, and versatility goals, the hardware accelerators have signifi­
                cantly improved final image legibility



Bitonal image technology, which can be viewed as            should come of age in the 1990s. Digital imaging
the electronic version of today's microfilm method ,        is a l ready in use in many areas and new applica­
is experiencing a high rate of growth . However, the        tions are being created for both com mercial and
electronic image data objects generated and manip­          scientific markets. The emergence of d igital images
u lated in this technology are very l arge and require      as standard data types supported by the majority
intensive processing. In a generic system, these            of systems (like text and graphics of today) seems
requirements can resul t in poor image processing           assured . For a greater understanding of specific
performance or reduced appl ication performance.            imaging appl ications, this section presents general
To address these needs, Digital has designed a series       imaging concepts and terms used throughout the
of imaging hardware accelerators for use in the doc­        paper.
ument m anagement market.
   This paper provides a brief tutorial on electronic       Concepts and Terms
imaging. It begins with a general description of the        In its simplest form, imaging is the d igital repre­
imaging data type and compares this type to the             sentation of real-world scenes or objects. Just as a
standard text and graphics data types. It continues         camera transfers a view of the real world onto a
with a discussion of specific issues in bitonal imag­       chemical film, an electronic imaging system trans­
ing, such as image data size, network transport             fers the same view i nto d igital bits for storage,
method, rendering speed , and end-user legibility.          manipulation, and v iewing. In this paper, the term
The paper then focuses on D igital's DECimage 1200          image refers to the digi tal bits and bytes that repre­
hardware accelerator for the VT 1200 X window               sent the real-world view.
terminal developed by the Video, Image and Print               The process of digitizing the v iew may be done
Systems Group. It concludes with fu ture image              through various methods, e.g., a n image scanner
accelerator demands for the processing of multi­            or image camera. A scanner is the conceptual
media applications and continuous-tone images.              inverse of a normal printer. A printer accepts an
                                                            electronic stream of bits that describe how to
Introduction to Imaging                                     place the ink on the paper to create the desired
Just as graphics technology blossomed in the 1980s,         p icture. Conversely, optical sensors in the scanner
electronic imaging and its associated technologies          transform l ight intensity values reflected from a



D igital Technical journal   Vol. 3 No. 4   Fa/11991                                                             9
Image Processing, Video Terminals, and Printer Technologies




sheet of paper and create a stream of electronic bits        media density. Scaling an image may be as simple as
to describe the p icture. Simi lar sensors in the focal      repl icating and dropping pixels, or i t may involve
plane of a camera produce the other common d igi­           i n terpola t ion and other algorithms that take neigh­
tization method, the electronic i mage camera.               boring pixels into account. Generally, the more
   The format of a digitized image has many param­           complex scaling algorithms require more process­
eters. A p ixel is the common name for a group of            ing power but yield higher-quality images, where
d igitized image bits that all correspond to the same       qua l i ty refers to how wel I the original scene is rep­
location in the image. This pixel contains informa­         resented in the resulting image.
tion about the intensity and color of the image a t             Before an image can be displayed, i ts pixel values
o n e location, in a format that c a n b e interpreted      often require conversion to account for the charac­
and transformed into a visible dot on a d isplay            teristics of the d isplay device. As a simple example,
device such as a printer or screen. The amoun t of          a color image cannot retain its color when output
information in the pixel classifies the image into          to a black-and-whi te video moni tor or printer. I n
one of three basic types.                                   general, when a device can d isplay fewer colors
                                                            than an image contains, the i mage p ixel values must
•    A bitonal image has only one b i t in each pixel;
                                                            be quantized . Simple quantizing, or thresholding,
     the bit is ei ther a one or a zero, representing one
                                                            can be used to reduce the number of image colors
     of two possible colors (usually black and white).
                                                            to the number of d isplay colors, but can resu lt in
•    A gray-scale image has multiple bits in each pixel,    loss of image quali ty. Di thering is a more sophisti­
     where each p ixel represents an intensity value        cated method of quantizing, which produces the
     between one color (al l zeros) and another color       il lusion of true gray scale or color. Although dither­
     (all ones). Since the two colors are usually black     ing need use no more colors than simple quantiz­
     and white, they produce a range of gray-scale          i ng, it results in d isplayed i mages of much higher
     values to represent the image.                         qual i ty.
                                                                Image compression is a transformation process
•    A color image has mu ltiple components per
                                                            used to reduce the amoun t of memory required to
     p ixel, where each component is a group of
                                                            store the information that represents the image.
     bits representing a value within a given range.
                                                            Different compression methods are used for bi tonal
     Each component of a color image corresponds
                                                            images than those used for gray-scale and color
     to a part of the color space in which it is repre­
                                                            images. These methods are standardized to specify
     sented. Color spaces may be thought of as dif­
                                                            exactly how to compress and decompress each
     ferent ways of representing the analog, visible
                                                            type of image. For bitonal images, the most com­
     range of colors in a d igitized, numeric form. The
                                                            mon standards are the ones used in facsimile
     most popular color spaces are television's YlN
                                                            machines, i.e. , Recommendations T.4 and T.6 of the
     format (one gray-scale and two color compo­
                                                            Comite Consultatif Internationale de Telegraphique
     nents) and the bit-mapped computer d isplay's
                                                            et Telephonique (CCJTT).1·2 Commonly known as
     RGB format (red, green, and blue components).
                                                            the Group 3 and Group 4 standards, the desig­
   The resolution of an image is simply the density         nations are often shortened to G3-I D , G3-2D, and
of pixels per unit distance; the most common den­           G4-2D, referring to the particular standard group
sities are measured i n dots per inch (dpi), where          aod to the coding method, which may be ei ther
a pixel is cal led a dot. For example, a facsimile          one- or two-dimensional. For gray-scale and color
machine (which is nothing more than a scanner,              images, the joint Photographic Experts Group
printer, and phone modem in the same unit) typi­            OPEG) standard is now emerging as a joint effort of
cally scans and prints at 100 dpi, although newer           the International Standards Organization (ISO) and
models are capable of up to 400 dpi. As another             CCITI.0 Whichever format or process is used, com­
example, most workstation d isplay monitors are             pression is a compute-intensive task that involves
capable of 75- to 100-dpi resolution, and some high­        mathematically removing redundancy from the
end monitors achieve up to 300-dpi reso lution.             pixel data.
   To d isplay an i m age a t a density different from          A typical compression method creates an
its scanned density, without altering the image's           encoded bit stream which cannot be d ispl ayed
original size, requi res the i m age to be scaled, so       directly; the compressed b i ts must be decom­
that the new image density matches the output               pressed before anything recognizable m ay be



10                                                                 Vol. 3 No. 4   Fu/119')1   Digital Techuicaljournal
                                                         Hardware Accelerators for Bitonal Image Processing



d isplayed. The term compression ratio represents              The major drawback in the imaging process is
the size of the original image divided by the size of      increased data size, which results in storage mem­
the compressed form. For bi tonal images using             ory and network transport problems. High scan
the CCITI standards, the ratio is commonly 20:1 on         densities and color information components create
normal paper documents, bu t can vary widely with          large volumes of data for each image; a bitonal
the actual content of the image. The CCITI stan­           image scanned at 300 dpi from an 8.5 -by-11-inch
dards are also " lossless" methods, which means that       sheet of paper requires over 1 megabyte of mem­
the decompressed image is guaranteed to be iden­           ory in its original pixel form. Therefore, compres­
tical to the original image (not one bit d ifferent).      sion and decompression are integral parts of any
In contrast, many " lossy" compression methods             imaging system. Even in compressed form, a bitonal
al low the user to vary the compression ratio such         image of a text page requires about 50 kilobytes
that a low ratio yields a nearly perfect image repro­      of storage, whereas its American standard code
duction and a high ratio yields a visible degradation      for information interchange (ASCII) text equivalent
in image qual i ty. This trade-off between compres­        requires only 4 to 5 kilobytes. Similarly, a graphics
sion and image qual i ty is very useful because of the     file to describe a simple block d iagram is much
wide range of applications in imaging. An applica­         smal ler than its scanned image equivalent.
tion need pay no more in memory space and band­                Based on these advantages and l imitations, sev­
width than necessary to meet image quality                  eral appl ications have emerged as perfect matches
requirements.                                              for imaging technology. Bitonal images are used
                                                           in the expanding market of docu ment manage­
A New Data Type and Its Features                           ment, which consists of scanning volumes of
The image data type is fundamentally different             papers into images. These images are stored and
from text and graphics. When a user views charac­          indexed for later searching and viewing. Basically
ters or pictures on a display device, the source of        an electronic file cabinet, this system resu lts in
that view is usually not important. A sheet of text        large savings in physical cabinet space, extremely
from a printer may have come from either a text file       fast document access, and the abi l i ty for mul tiple
where the printer's own fonts were used, a graph­           users to access the same document simu ltaneously.
ics file where the characters were drawn with line         Gray-scale imaging i s often used in medical appli­
primitives, or an image file where the original text       cations. Electronic versions of X rays can be sent
document was scanned i nto the system. In any              instantly to any special ist in the world for d iagno­
case, the same letters and words present the user          sis, and the ordering of sequential computer-aided
with approx imately the same information; the dif­         testing (CAT)-scan images into a "volume" can pro­
ferences are mostly in character qual ity and format.      vide valuable three-dimensional views. The appli­
   In spite of their large storage space require­          cations for color imaging are relatively new and
ments, images have several advantages over graph­           still emerging, but some are already in use commer­
ics or text. F irst, consider the process of getting       cially, e.g., l icense and conference registration pho­
the information into the computer. With the imag­           tographs. A fu rther extension to still imaging is
ing process, documents may be scanned automati­             digital video, which can be considered as a stream
cally in a few seconds or less, compared to the time        of still images. In conjunction with audio, d igital
required for someone to type the information cor­          video is commonly known as m u ltimedia, appl ica­
rectly (absolutely no errors) into a text file. Also,       tions for which range from promotional presenta­
even though the software exists to convert elec­            tions to a manufacturing assembly process tutorial.
tronic raster images into graphic primitive files, the         In this paper, we focus on the static bitonal imag­
process loses detail from the original image and is        ing method of representing real-world data inside
relatively slow. Next, consider the variety of infor­      computers. Static imaging is a simpler method of
mation possible on a sheet of paper: a user can­           representing a broader range of information than
not easily reproduce a diagram or a signature on a         the text and graphics media types, but i t carries
document. A scanned image preserves not only the           a greater requirement for processing power and
characters, but their font, size, boldness, relative        memory space. In addition, static imaging can be
position, any pictures on the page, and even               viewed as one part of true multimedia, as can text,
smudges or tears depending on the quality of the            graphics, audio, video, and any other media for­
image scan.                                                 mats. Yet static imaging does not have the system



D igital Technical journal   Vol. 3 No. 4   Fall 1991                                                          11
I mage Processing, Video Terminals, and Printer Technologies




speed requirements of a motion v ideo and a u d i o          m ately 3 k ilobytes of memory. If the same sheet of
system, which mu st present d a t a at real - time rates.    paper is d ig i tized by scan n i ng at various dot den­
As long as the user can tlcal with static i magcs at         sities, the result ing data tiles are huge, as shown
an i n teractive rate, i . e . , being ablc to v iew the     by the decom pressed bitonal im age sizes in Table ! .
images in the fo rmat of choice as fast as the user          Note t ha t Table 2 i ncludes the size of the scanned
can sdect them, then static i m aging is a powerful          i m age if scan ned in gray-scale and color modes,
media presentation too l. The next section presents          alt hough using these modes would not make sense
the important issues concerning bitonal imaging in           on a black-and-whit�: sheet of paper. The image
a tlocumcnt managcment env ironment.                         sizes arc included for comparison and are d iscussed
                                                             in the section Future I mage Accelerator Requi re­
Bitonal Imaging Issues                                       ments. The clara presented in Tables 1 and 2 i l l us­
As previously mentioned, b i tonal electro n i c i m ag­     trates that the size of the original ASCII file is much
ing as an alternative to paper documcnts offers              smal ler than any of the scanned versions. The data
many benehts, such as reducctl physical storage              also gives evidence that scanned images, in general,
space, instant and simu ltaneous access of scanned           require considerable memory.
images, and in general a more accessible media.                Since the typical use for b i tonal images is for
Serious issues need to be resolvetl before a produc­         vol u m e document archival, an imaging application
t ive i m aging operation can be implemented. The            must include a compression process to reduce mem­
chief issues are the image data size, transport              ory usage. This process must transform the original
method , perceivetl rendering speed, and final legi­         scanned image file to a much smaller file without
b i l ity. In the fol lowing sections, we examine each       losing the content of the original scanned data.
issue and present sol u tions.                                 Compression algorithms may take d ifferent paths
                                                             to achieve the same result, but they share one basic
Digitized Image Data Size                                    process, the removal of red u n d a n t i nformation to
The most important issue concerns image data size.           reduce the obj ect size. A common compression
I mages are typically documents, drawings, or p ic­          rout i n e searches the p ixel data for groupings, or
tures that have been d igitized i nto a computer­            "run lengths," of black or white pixels. Each run
readable form for storage antl retrieval. Depending          length is assigned a code sign ificantly shorter than
o n the dot densi ty of the scanner, a s i ngle i m age      the ru n length i tself The codes are assigned by
can be I to 30 megabytes or more in size. However,           sta tistics, where the most frequent run lengths
storing a single image in i t s scanned form is not the      are assigned the shortest codes; statistics have been
typical usage model . I nstead , a company may have          amassed on a variety of document types for d iffer­
tens of thousands of scanned documents. Clearly,             e n t scan densi ties and document sizes. A compres­
with today's storage technologies, a company can­            sion process p arses through the origi nal im age
not af ord to store such a large vol u m e of i mages        file, genera t i ng another tile that contains the codes
in that format.                                              representing the origi nal image. Figure I , a sample
     A typical ASC I I file represen t i ng the text on an   b i t o n a l i m age co mpression, i l lustrates these com­
8.5 -by-1 1-inch sheet of paper requires approxi-            pressed codes in a serial bi t stream.



Table 1      Sa m ple Bitonal I mage Sizes

                                       Scan                                         Kilol'ytes of Data
Document Type                          Density                  Pixel Form                                Ty pical
(Paper Size)                           (dpi)                    (Decom pressed)                            Compressed

A size                                   1 00                            114                                     46
(8.5 X 11 i nch)                         200                              457                                    47
                                         300                            1 027                                    50

E si ze                                  1 00                           1 826                                   1 06
(44 x 34 i n ch)                         200                            7305                                    114
                                         300                          1 6436                                    1 27




12                                                                   Vol. 3 No. 4   Fall 1991   Digital Technical journal
                                                                                          Hardware Acceleratorsfor Bitonal Image Processing



Table 2     Sample Gray-scale a n d Color                                                      Network Transport Constraints
            I mage Sizes                                                                      The network transport performance for an image
                                                   Kilobytes of Data
                                                                                              is important, because images arc most often stored
Document Ty pe                                     in Pixel Form                              on a remote system and v iewed on a widespread
and Size                                           (Decom pressed)                            group of d isplay stations. For example, one group in
                                                                                              an insurance company receives and scans claim
1 28 x 1 28 pixel, 1 2 bits per pixel                               24
                                                                                              papers to create a central ized image database,
g ray-scale i m age
                                                                                              while users in another group acn:ss the documents
51 2 X 512 pixel, 8 bits per pixel                                256
color i mage
                                                                                              simultaneously to process clai ms. For the imaging
                                                                  768
                                                                                              system to be productive. this image data needs
51 2 x 51 2 pixel, 24 bits per pixel
color image                                                                                   to be transported quickly from one group to the
8.5 X 11 i nch, 100 d p i , 24 bits                            2740                           other: telephone attendants answering calls must
per pixel, color i m age                                                                      have im mediate access to the data.
                                                                                                 Scanned image documents take a long time to
                                                                                              transport between systems, simply because they are
   Several algorithms for bitonal compression are                                             so large. When compression techniques are used , a
widdy used today. As mentioned in the previous                                                typical u ncompressed image stored in 1 megabyte
section, the most common for bitonal images are                                               can be reduced to approxima tely 50 kilobytes.
the CCllT standards G3-l D, G3-2D, and G4-2D, which                                           Since transport time is proportional to the number
all use the approach just described. For the one­                                             of packets that must be sent across the network,
dimensional method, the algorithm creates run                                                 reducing the data size to .:; percent of its original
lengths from all pixels on the same scan l ine. In the                                        size also reduces the transport time to 5 percent
two-dimensional methods, the algori thm some­                                                 of the original time. Therefore, you can now send
times creates run lengths the same way, but the                                               twenty compressed images in the same time previ­
previous scan l ine is a lso examined. Some codes                                             ously spent sending one uncompressed image.
represent run lengths and even whole scan l ines                                                  Even with compression techniques, the image
as " the same as the one in the previous scan line,                                           files are stil l larger than their text file equivalents.
except offset by .v pixels," where N is a small inte­                                         Moreover, most network protocols limit their
ger. The two-dimensional method takes advan­                                                  packet size to a maximum number of bytes, i.e . , an
tage of most of the redundancy in an image and                                                image file larger than the maximum packet size
returns the smallest compressed file. In addition to                                          gets divided over mul tiple packets. If the protocol
preserving system memory, these compression                                                   requires an acknowledgmenr between packets, then
methods significantly improve network transport                                               the transport of a large file over a busy network
performance.                                                                                  becomes a lengthy operation.


                                IMAGE PI XELS




                                                                                      r
                                t?t �� �U
                                                                        AN LI N E o F '



                                                          1 r---
                                                                      o                     Ls



                                                            1
                                                                        -   _           -
                                                                     _�            __,...,.    --,




                                                                   600 W H I TE                250 WHITE                  845 WHITE
                                                                                                                     ACK




                      . . . 0 1 1 0 1 0000 1 0 1 000 00000 1 1 0 1 000 0 1 0 1 1 1 0 1 0 1 1 0 1 1   00000 1 1 0 1 000 0 1 1 0 1 00 1 00000 1 1 . .

                                            C O M P R E S S E D C O D E S I N A CONTINUOUS B I T STREAM




                                                     Figure I            Bitonal Image Compression



Digital Techllical Jour11al         VlJI. 3 No. 4         Fall 1991                                                                                   13
I mage Processing, Video Terminals, and Printer Technologies




   The platform for our most recent accelerator is                  them. The disk acce ss time data from the DECim age
the VT1200 X window term inal, which uses the                       project a n al ysis shown in Table 3 d emonstrates that
l ocal are a transport ( I.AT) network protoco l . We               the disk file re ad time is a significant portion of the
soon realized that the X server packet size w as                   overal l response time. Thus, the display station
l imited to 16 kilobytes and the t yp ical A-size                  render time is the onl y are a of the disp l ay response
compressed docu ment was approximatel y 50 kilo­                    time which can be clear l y intluenced and is, there­
bytes. With this arrangement, each i m age transport               fore, the m ai n focus of our image accelerators. The
woul d have required four large data packets and                   local processing that must occur at the display sta­
four acknowledgment p ackets. Working with the                     tion is not a trivial t ask; an im age must be decom­
X Window Terminal B ase System Software Group,                     pressed, scaled, and clipped to fit the user's current
we were able to r ai se the packet size limit to                   window size, and optional l y rotated.
64 ki lobytes. The b ase s ystem group also imple­                     The decompression procedure inverts the com­
mented a delayed acknowledgment scheme, which                      pression process; both are computational l y com­
eliminates the need for the c l ient to wait for an                p lex . Input to the procedure is compressed data,
acknowledgment packet before sending the next                      and output is the origi n al scan l ine p ixel data,
d at a packet. Table 3 shows compressed image d at a               which can be written to a disp l ay device. Scal ing
taken during the DECim age 1200 d evelopment                       the data to fit the current window or fill a region
c ycle. Notice that the network transport t imes                   of interest is not trivi al either: a huge inpu t dat a
for Digi t al docu ment interch ange format (DDIF)                 stre am must b e processed (the decompressed, orig­
decrease sharpl y after the p acket changes.                       inal file), and a moderate output data stream must
                                                                   be cre ated (the viewable image to be displ ayed).
Perceived Rendering Speed                                          While simple pixel replicate and drop algorithms
Because the imag e scanning and compression                        m ay be used to scale the d a t a, a more sophisticated
operations occur only once, they are not as                        scaling algorithm h as been shown to grea t l y
perform ance-cri tical as the decompression and                    enh ance the output i m age qual i t y.
rendering for displ ay operations, which are done                      In add i tion to scal ing and cl ipping, the orthogo­
m any times. Decompression and rendering are p art                 n al rotation of image s (in 90-degree increments)
of the s ystem's disp l ay response time, which is a               is a useful function on a display station. Some docu­
critical factor in a system designed for high-volume               ments m a y h ave words running in one direction
appl ications that access thousands of i m ages dai ly.            while pictures are oriented another way, or the user
This time is measured from the inst an t the user                  m ay wish to view a portrait-mode image in l and­
presses the ke y to s elect an im age to view, to the              scape mode. In either case, orthogonal rotation
moment the image is displayed completely on the                    c an help the user understand the inform ation; i.e . ,
screen. The disp l ay response time is a fu nction of              the increased time to ro t ate the view is warranted.
the disk read time, network transport time, and dis­                   When an image is scan ned, p articu l arl y with a
play st ation render time.                                         h and-held scanner, the p aper is never perfectl y
  Although network transport time and disk file                    al igned. Thus, the image often requires a rotation of
read time have a direct effect on the response time,               I to 10 degrees to m ake the view appear straight
accelerator developers rarel y have any control over               in the i m age file. However, m u l tiple users want the


Table 3       DD I F I mage F i le Read Ti me a n d F i l e Tra nsport Perfo rma nce

                                                                                                 Network Tra nsport Time
                                                                                                         (mi l l iseconds)
                                             Disk Read Time                                      After                 Before
I mage Size                                   (m i l l iseconds)                                 Packet                Packet
(kilobytes)              M i c roVAX I I        VAX 8800                   VAX 6440              Change                Change

      19                     1 223                   480                      281                     325                    960

      41                     1 5 34                  655                      332                     61 4               1 792

      99                     2351                   1 035                     598                    1 351               3928

     1 57                    3288                   1 380                     71 6                   2283                6430




14                                                                        Vol. 3 No. 4   Fall 1991   Digital Teclmicaljourual
                                                           Hardware Accelerators for Bitonal Image Processing



information from the document as quickly as pos­             General Design Strategy
sible , and should not have to ro tate the image by          The nu mber of applications using bitonal i mage
a few degrees to make it perfectly straight on the           data continues to increase. In general, these appli­
screen. Therefore, this m inimal ro tation should            cations attempt to offer low cost while achieving
be done after the initial scanning process; i.e.,            an interactive level of performance, defined as
only once, prior to indexing the material into               no more than 1 second from point of request to
the database, and not by every user in a distributed         complete image display. Ultimately, software may
environment. Because any form of rotation is                 provide this fu nctio nality without hardware accel­
compute-intensive, allowing the user to perfo rm             eration, but today's software cannot. Mo reover, the
minimal ro tations at a h igh-vol u me view station          parameters of image systems are n o t static; scan
wo uld reduce the application's perceived ren­               densities, overall image size, and the number of
dering speed and add l i ttle value to the station 's        images per database will a l l increase. These
function.                                                    increases will provide the most incentive for hard­
                                                             ware assist at the low end of the X window ter­
Final Legibility                                             minals market, because software alone cannot
While the primary issue facing im aging applica­             perform the amount of processing that users will
tions is data size, image viewing issues must also be        expect for their investment.
addressed. In short, an effective b i tonal imaging
                                                             The User Model Although a single mo del cannot
display system must be responsive to overal l image
                                                             suit every application, imaging is centered o n cer­
display performance and the resulting quality of
                                                             tain functions. Therefore, a user mo del built on
the image displayed. To enhance our products, we
                                                             these fu nctions wou ld be very useful in mapping
optimized the display perfo rmance parameters as
                                                             individual steps to the hardware: hardware versus
best we could, given that some parameters are not
                                                             software performance, the fu nction 's frequency of
under our control. Improvements to m o nitor reso­
                                                             use, and the cost of implementation.
lution and scanner densit ies continue to increase
                                                                The general user model for bitonal imaging sys­
the legibility of images. An affordable image system
                                                             tems is relatively simple. A small market exists fo r
should increase the image legibility by rendering
                                                             image entry stations, i n which documents are
a bitonal image into a gray-scale image using stan­
                                                             scanned , ed i ted, and indexed into a database. While
dard image processing techniques. We discuss the
                                                             a high throughput rate is important at these sta­
method used in our accelerators, i.e., an intel ligent
                                                             tions, a general-purpose image accelerator is not
scale operation in the hardware pipel i ne , in the
                                                             the solution-dedicated entry stations already
next section.
                                                             exist in the market. Instead, we designed a general­
                                                             purpose platform, or versatile media view station,
Hardware Accelerator Design                                  to be used for imaging appl ications alongside other
As explained i n the previous section, transforming          applications. The user model fo r this larger market
documents into a stream of electronic bits is not            is a set of operations for viewing and manipu lating
the demanding part of a bitonal imaging process              images al ready entered i nto a database. The most
for document management. Also, scanners and                  common operations in this mo del are decompres­
dedicated i mage data-entry stations abound in the           sion, scal ing, clipping, orthogonal ro tation, and
marketplace al ready. Instead, the challenge l ies i n :     region-of-interest zooming.
( 1 ) managing t h e image data size t o control
memory costs and reduce network slowdown;                    Display Performance and Quality Optimization
(2) increasing the i mage rendering speed, i.e.,             The main thrust of the DECimage accelerator is to
decompress the image , scale it, and clip i t to fit         achieve interactive perfo rmance for the operations
the window size with optional rotation; and                  defined in the user model. A secondary goal is to
(3) increasing the quality of the displayed images.          bring added value to the system by increasing the
This section describes the way our strategy                  qual ity of the displayed image compared to the
i nfluenced the design of DECimage products. We              quality of the scanned image. A side effect of maxi­
also discuss the chips used for deco mpression and           mizing performance in hardware is that the main
scaling, and how Digital's existing client-server pro­       system processor has wo rk off-lo aded fro m it, free­
tocols support these imaging hardware accelerators.          ing it for other tasks.




Digital Technical ]oun1al   Vol. 3 No. 4   Fall 1991                                                            15
Image Processing, Video Te rminals, and Printer Tech nologies




     The general design of the accelerator uses a               The most straightforwa rd method o f i n t egra tion is
pipel ined approach. Since maximum performance                  to relocate the hardware from the present option
is desired and a large amount of data must be pro­              to the m a i n system processor board; successive
cessed by the accelerator board, m u l t iple passes            steps of i ntegra tion wou ld consol idate m apped
through the board arc not feasible. Simi larly, the tar­        hardware to fewer total devices. The most cost­
geted low cost does not a llow a whole image bu ffer            effective i ntegra tion wil l be the inclusion of the
on the board. With one exception (rota ti on), a l l            mapped hardware in the processor in a way simi lar
board processing shou ld b e dune i n one pipel i ne,           to a float ing-point u n i t (FI'l"). Just as gra phics accel­
with the system processor simply feeding the i n p u t          eration is now being i ncl uded in system processor
e n d of t h e pipe and dra i n i ng t h e o u t p u t encl .   design, images w i l l eventu a l ly achieve the status of
lkcause of t h e large amount of d a t a t o b e read from      a requ ired data type and thus be supported in the
the board and displayed on the screen , the proces­             base system p rocessor.
sor should o n ly have to move that data, not do any
further opera t ions on i t. lo this end, any logic             Product Definition- What Does the User
requ i red to format the p ixels for the display b i t map      Want?
shou ld be included in the pipeline.                            The previously descri bed strategy was u sed in the
                                                                design of the image accell"rator board for the
Cost Reduction through Less E\pensive System                    DECimage 1200 system. The product req u i rements
Components The net cost of a bitonal imaging                    cal led for a low-cost, high-performance document
system is i nfluenced by the capab i l i t y of the assist      i mage view statio n . These req u i rements evolved
hardware. The capabil ity of the hardware impl ies              from the belief t h a t most users currently i nvesti­
Oexibil i t y in the choice of other system hardware.           gating imaging systems are i n terested i n appl ica­
In t h is regard, the most significant impact o n cost          tions and hardware that will enable them to quickly
occurs in the memory and the display. A system that             and s i m u l taneously view document im ages and run
makes use of fast decompression and sca l i ng hard­            their existing non imaging appl ica tions. These users
ware can quickly d isplay compressed im ages from               are involved with com mercial and business appl i­
memory. This means either more images can be                    cations, rather than scie n tific app l i ca t ions. The
maintai ned in the same memory, or the system can               DECimage 1200 system was planned for the manage­
operate with less memory than i t wou ld witho u t              ment of insurance claims processing,                 hosp i tal
the assist hardware; less memory means lower cost.              pa tient medical records, bank records, a nd manu­
     A mon.: dramatic effect on system cost is in the           facturing documents. As previously stated, the
d isplay. Imaging systems general ly need higher­               i m aging fu nctions requ i red for these view-oriented
density displays than n o nimaging systems, but the             appl ications are high-speed decompression, scal­
cost of a 150-dpi disp l ay is approximately twice the          i ng, rotation, zooming, and clipping.
cost of a 100 - d p i display of the same d imensio ns.
However, we found that we could i ncrease legi b i l­           General Product Design
ity, i .e . , expand a bitonal im age to a gray-scale repre­
                                                                In defining the image capable system, the key
senta tion, by using an intel l igent sca le opera t i on
                                                                points in the product requirements l ist were
in the hardware pipeline. For example, a bitonal
i mage rendered to a 100-dpi d isplay using the i ntel­         •   High-performance image display
l igent scale process gives the perceived legi b i l i t y
                                                                •   Low cost
of the same i m age rendered to a 150 - d p i di splay
with a simple sca l i ng method. That is, hy adding the         •   Bitonal images only (not gray-scale or color)
i n te l l igent scale, a 100-dp i d isplay can be used
                                                                •   View-o n ly functions
where prev iously o nly a 150-dpi di splay wou ld be
adequate.                                                           The need for h igh-performance display influ­
                                                                enced the project team to design the hardware
Cost Reduction through In tegration           Presently, as     accelerator board to handle im age decom press ion,
in the DECimage 1200, hardware-assisted image                   scali ng , a nd ro tation. Previous performance test­
manipulation exists as a board-ll"vel option. Higher            ing on a 3-VUP (VAX-1 1/780 u n i ts of p erformance)
levels of i ntegra tion with the base p l a t form w i l l      CPU had yie·lded image software disp lay t i mes from
provide lower overa l l cost for an imaging system.             5 to 19 seconds. These images were compressed




16                                                                      lk)f. 3 No. 4   rail 19'JI   Digital Techt�ical journal
                                                                         Hardware Acceleratorsfor Bitonal Image Processing



according to the cenT Group 4 standard (300 dpi,                           Group 3 and Group 4 bitonal compression algo­
8.5 -by-1 1 inches), and ranged from 20 to 100 kilo­                       rithms. This specialization al lowed the use of a
bytes in size. In addition, the software display times                     Digital application-specific integrated circu it (ASIC)
were highly dependent on the image data content.                           decompression chip. Finally, the view-only require­
The more complex image files, which had lower                              ment limited the scope and complexity of the
compression ratios, took significantly longer to                           design by eliminating the need for extra hardware
decompress, scale, and display than the simpler                            to hand le the compression of im ages after they
image files. For example, an A-size, 300 -dpi, cenT                        have been scanned and edited .
Group 4 compressed image w i th a compression
ratio of 10:1 took approximately 18 seconds to
                                                                           Specific Product Design
d isplay, while another with a ratio of 3 3 : 1 took
approximately 7 seconds.                                                   The decisions described in the previous section led
   The other three requi rements Jed to decisions                          to our design of an image accelerator board that
about the specific design of the image accelerator                         supports: CCI1T Group 3 and Group 4 image decom­
board. The need for low cost meant designing an                            pression using an ASIC decompression chip; integer
option for an existing low-cost platform, which led                        scaling using an ASIC scaling chi p ; orthogonal rota­
us to D igital's VT1200 X window terminal. This                            tion; and image d isplay. Figure 3 shows a general
requ irement also led to our support of the pro­                           block d i agram of the board and how it fits into
posed X Image Extension (XIE) protocol.4 The XIE                           Digital's VT1200 system architecture. The accelera­
protocol extends the X l l core protocol to enable                         tor board is a ttached to the system address/data
the transfer of compressed images across the wire                          bus, and its registers, data input port, and data
and to enable interactive image rend i tion and d is­                      output port are m apped into the C PU s 110 space.
                                                                                                                    '




play at the server. In the X windowing client-server                       The accelerator board is accessed by reading and
environment, image applications and compressed                             writing specific addresses l ike any other system
im age files exist on the client host machine, as                          memory space . Note that the image accelerator
depicted in Figure 2. In addition, the XIE protocol                        logic is separate from the video terminal logic.
standardizes the interface-to-image functions in                           Decompressed images are read from the image
the X windowing environment and enables the                                board and writ ten to the base system video mem­
development of a common appl ication that can be                           ory for d isplay.
used on any XIE-capable station. The client appl ica­                         The main operation consists of the fol lowing
tion issues commands to the X server display sub­                          steps: compressed image data is read from system
system and the XIE special ized image subsystem .                          memory and wri tten to the ASIC decompression
When a user selects an i mage to view, the com­                            bu ffer by the processor; the data is then decom­
pressed image file is transported from the client­                         pressed, scaled by the ASIC seal ing chip, packed
side storage device to the X server memory.                                into words, and written to the out p u t buffer.
    Because the proposed accelerator would han­                            Figure 4 shows a detailed block d iagram of the
d le only bitonal images, we could special ize our                         image accelerator board logic. The scal ing chip out­
board to decompress only the standard CCilT                                puts p ixels of data (1 bit per p ixel in this case)
                                                                           which are packed into words using shift registers.
                                                                           As soon as a word of data is available, the scaling
                                                                           chip output halts. Control signals generated in pro­
      X/X I E C L I E NT                           X/X I E
      APPLICATION                                  SERVER
                                                                           grammable array logic (PAL) write the packed word
                            X 1 1 AND XIE
                            WIRE
                                                                           into the output buffer and tell the scal ing chip to



              I
                            PROTOCOL                                       begin outpu t ting pixels again. When the out p u t
                                                                           buffer is ful l , the processor reads the rendered
                                                                           image data from the buffer. If rotation is requ ired,
  COMPRESSED                                                               the processor writes the data to the rotation
  IMAGE STORAGE                                            IMAGE           matrix; otherwise, the data is cl ipped and wri tten
                                            D ISPLAY
  • D ISK                                                  H A R DWARE
  •    C D ROM                                                             to the bit map. The image driver software, after set­
                                                                           ting up the board, al ternates between checking
                                                                           whether the input bu ffer is empty and whether the
           Figure 2        X Client-server Architecture                    output buffer is fu ll.




D igital Techllicaljournal          Vol. 3 No. 4       Fall 1991                                                               17
Image Processing, Video Terminals, and Printer Tech nologies




                         c:J
                                                NETWORK              SYSTEM                   IMAGE
                                                INTERFACE            MEMORY                   HARDWARE



                                  t                        !             t                        !
                       <                             SYSTEM ADDRESS/DATA BUS
                                                                                                          >
                                  f                        !             !
                             OPTION             DIAGNOSTIC           VIDEO
                             MEMORY             ROM                  MEMORY          f-        DISPLAY




                                         F(!{ure 3     VT/200 System Architecture


   The rotation circu it handles 90- and 270-degree                to be decompressed before it can begin scaling;
rota tion, whereas 180-clegree rotation is hand led                insteacl, scal ing begins as soon as the first byte of
in the data packing shift registers by changing the                data is ou tput from the decom pressor. Thus dif­
shift direction. The circu it rotates an 8-by-8-bit                ferent pieces of the image fi le are being decom­
bl ock of data at a time. The fi rst byte of eight con­            pressed, scaled, and rotated simu ltaneously. The
secut ive scan l ines is wri t ten into eight individual           hardware pipeline also eliminates the need to
byte-wide registers. The most significant bit (MSB)                store the ful ly uncom pressed image (approximately
of each of these registers is connected to the byte­                1 megabyte of data for A-size 300 -dpi im ages) in
wide rotation output port latch. A processor read                  memory. The compressed image is written from
of this port triggers a simultaneous shift in a l l of the         system memory to the accelerator board and a
rota tion data registers so that the next bit of each              decompressed, scaled, and clipped image is read
register is now l atched at the rotation output port               from the board . Because of the speed of the hard­
for the next read. Figure 5 diagrams the rotation                  ware, the software can redisplay an i mage with dif­
circuitry just described.                                          ferent seal ing, cJ ipping, or ro tation parameters; i t
   lb achieve the best performance, we p ipel ined                 merely changes t h e hardware setup for t h e differ­
the functional blocks in the hardware. The scal ing                ent parameters and sends the compressed image
engine docs not need to wai t for the entire image                 tile back through the accelerator board pi pel in e.



                           IN PUT              SCAN LINE
                           BUFFER              RAM B U FFER



                              !                        !
                      DECOMPRESSOR               SCALING           DATA PACKING               FIFO OUTPUT
                      CHIP         -             CHIP          r-- CIRCUIT           - BUFFER



                              !                  SYSTEM ADDRESS/DATA BUS
                                                                                                   t

                                        COMMAND
                                                I                              t
                                                                               8x8
                                        AND STATUS                             ROTATION
                                        REGISTER                               MATR IX



                      Figure 4        Block Diagram of DECimage 1200 Accelemtor Hardware



18                                                                           Vol. . i No. 4   Fall 1991   Digital Technical journal
                                                                                         Hardware Acceleratorsj01· Bitonal /mage Processing



                     8 SCAN LINES OF IMAGE DATA                                            the word contain the number of consecutive white
 8-BYTE 1 8-BYTE 1 8-BYTE


                                      I
                                                                                           or black pixels (called the run length). The upper
 BLOCK    BLOCK    BLOCK
 #1       #2       #3                                                                      bit of the word contains the run-length color code
    I                                                                                      (0 for a white run and 1 for a black run). An FLC is
                                                                                           read from the FIFO buffer and decoded into one of
                                                                                           eight routine types. Each routine is made up of sev­
                          8x8                                ROTATION
                          ROTATION                           OUTPUT                        eral states that control the color code toggling, run­
                          M AT R I X                         REGISTER
                                                                                           length adder, and accumu lator circuits. At the end
                 0    1   2   3   4       5   6   7                                        of each routine, a new word containing the run­
SCAN LINE 0      t---1                                                                     length and color information is written into a FIFO
SCAN LINE 1
                                                                                           buffer for the final stage.
SCAN L I N E 2
                 t---1
SCAN LINE 3 1---i -                                                                           The final stage of the decompressor chip con­
SCAN L I N E 4
            1---t                                                                          verts the run-length and color information to black
SCAN LINE 5 1---i                                                                          or whi te pixels. This stage outputs these pixels in
SCAN LINE 6 1---i
                                                                                            16-bit chunks when the scaling chip sends a signal
SCAN LINE 7
                 '---'
                                                                                           indicating a readiness to accept more data.

                                                       0     1   2   3   4   5   6   7
                                                                                           Scaling Chip     The primary purpose of the scal ing
                                                             D I S P LAY BYTE              chip is to input h igh-resolution document images
                                                                                           (300 dpi) and scale them for display on a medium­
                                                                                           density moni tor ( 100 dpi). The chip offers inde­
                                                                                           pendent scaling in the horizontal and vertical
                 Figure 5             Rotation Matrix
                                                                                           d i rections. The scaling design implemented in the
                                                                                           chip is a patented algorithm that maps the input
ASIC Design Description                                                                    image space to the output image space. General
The ASIC design consists of a decompressor chip,                                           M-to-N pixel scaling is provided where M and N are
which decodes the compressed image data to pixel                                           integers between 1 and 127, with the delta between
image data, and a scal ing chip, which converts the                                        them less than 65. M represents the number of pix­
image from the input size to the desired d isplay size .                                   els in and N represents the number of p ixels out (in
                                                                                           the approximated scale factor) .
Decompressor Chip          The decompressor chip acts                                          Given an image inpu t size and a desired d isplay
as a CCITT binary image decoder. The chip contains                                         size, we must find the M and N scale factors that
three distinct stages, which are pipel ined for the                                        best approximate the desired scale factor, wi thin
most efficient data processing. Double buffering                                           the range l imits of M and N as previously stated.
of compressed input data is implemented to enable                                          Thus an input width of 3300 and a desired out­
simultaneous input data loading and image decod­                                           put width of 550 are represented by an 111 of 6 and
ing to occur. Compressed data is loaded into the                                           an N of 1. The approximated M and N values are
input buffer by the processor through a 16- or 32-bit                                      loaded into the chip scale registers for downscaling
port. Handshaking controls the transfer of decom­                                          or upscal i ng.
pressed data from the decompressor's 8-bit-wide                                                The chip scaling logic uses the scale register val­
output bus to the scal ing chip.                                                           ues to increment the input pointer position and
   The first stage of the decompressor chip con­                                           generate output pixels. A la tched increment deci­
verts CCITT-standard Huffman codes, which are of                                           sion term is updated every clock cycle, based on
variable-length, to 8-bit, fixed-length codes (FLCs) 5                                     the previous term and the scale register values.
A sequential tree fol l ower circu i t is implemented                                      When scal ing down (where fewer pixels are outpu t
to handle this conversion. Every Huffman code cor­                                         than are input), the logic increments the input
responds to a unique path through the tree, which                                          pointer position every clock cycle, but on ly out­
ends at a leaf indicating the FLC. The 8-bit FLC is                                         puts a pixel when the increment decision term is
sent to a first-in, first-out (FIFO) buffer, which holds                                   greater than or equal to zero. Figure 6a il lustrates
the data for the second stage.                                                             how this algorithm maps input pixels to output pix­
   The second stage of the chip generates a 16 -bit,                                        els for a sample reduction. When scal ing up (where
run- length value from the FLC. The lower 15 bits of                                       every input pixel represents at least one output



Digital Technical journal                     Vol. 3 No. 4       Fall 1991                                                                     19
Image Processing, Video Terminals, and Printer Technologies




                    SCALE DOWN FROM 10 I N PUT PIXELS TO 5 OUTPUT PIXELS
                    ( M 2 AND N 1 )
                        =                =




                     I NIT = N 1 =


                     DELTA1 2N 2 =           =


                     DELTA2 = 2N - 2M                    = -   2

                                                               2            3           4           5           6           7          8             9             10
                    I NPUT                   •                 •           0            0           •           •           0          •             •             •


                    INCREMENT
                    DECISION  D              =   1           D = -1 D       =   1   D   = -   1 D   =       1 D = -1 D       =   1 D   = -     1 D   =       1 D   = -   1
                    REGISTER

                   OUTPUT                    •                             0                        •                       0                       •
                                             1                             2                        3                       4                        5



                                                                                (a) Downscaling



                    SCALE U P FROM 3 INPUT PIXELS TO 9 OUTPUT PIX ELS
                   (M  1 AND N 3)
                        =                =




                     I N IT 2M N = -1
                             =       -


                     DELTA1 = 2M = 2
                     DELTA2 2M 2N=           -           =   -4


                                                                                                        2                                  3
                   I NPUT                                        •                                  •                                      •


                   I NCREMENT
                   DECISION              D   = -     1       D     =   1   D -3 D
                                                                            =           =   -1 D        =   1   D   =   -3 D = -1      D   =    1   D -3 =


                   REGISTER

                   OUTPUT                    •                   •          •           0           •               0           0          0             •
                                                                  2         3           4           5               6           7          8             9



                                                                                (b) Upscaling


                                                             Figure 6               Chip Scaling Examples


pixel), the.: logic outputs a pixel every clock cycle,                                                  tion and display at the server using the hardware
but only increments the input pointer position                                                          accelerator board. Like the X l l protocol, the XIE
when the increment decision term is greater than                                                        protocol consists of a client-side l ibrary called
or equal to zero. Figure 6b il lustrates how this algo­                                                 XIElib, which provides cl ient applications access
rithm maps input pixels to output pixels for a sam­                                                     to image routines, and a server-side piece, wh ich
ple magn i.fication. For both cases, the value of the                                                   executes the cl ient requests. The X I E server imple­
pixel (black or w h i te) being output is t he value of                                                 ments support at two levels: device-independent
the input pixel pointed at du ring that clock cycle.                                                    and device-dependen t. The device-dependent level
In this dc.:scription, simply substitute rows for pix­                                                  supports the fu nctions that benefit from optimi­
els to repn:sent the vertical scaling process.                                                          zation for a particular platform , or functions that
                                                                                                        are implemented i n hardware accelerators. The
Software Supportfor the Hardware                                                                        device-independent level enables qu ick porting of
Software support is needed to enhance the func­                                                         functional ity from platform to pla tform. Figure 7
tions of the hardware accelerator in our image view                                                     illustrates the X/XIE cl ient-server architecture.
station. As ment ioned in the section General                                                              The c l ient-s ide X I Ei i b offers the minimum
Product Design, the XI E protocol extends the X l l                                                     functions necessary for im age re ndition and dis­
core protocol to enable the transfer of compressed                                                      play. The tool kit level offers higher-level rou tines
images across the wire and to enable image rendi-                                                       that assist with windows application development.



20                                                                                                                      11JI. 3 No. 4          Fall 19';)1         Digital Technical journal
                                                                Hardware Acceleratorsfor Bitonal /mage Processing



                                                                  which is essentially as fast as the user can ask for
                           APPLICATIONS                           the d isplays. This speed is possible because the
                                                                  image already resides in compressed form in the
                                                                  server memory. Thus, the image does not have to
                                                                  be read from the disk or transported across the
                                                                  network.




                             +
                                                                  Future Image Accelerator

                  XLIB                    XIELIB        I         Requirements
                                                                  Hardware accelerators will continue to be required
                                                                  for bi tonal imaging until software can provide the
                                                                  same functional ity at the same performance level.
                                                                  This section d iscusses the more complex image
                                                                  schemes that are used for gray-scale imaging and
                                  DEVICE INDEPENDENT ( D I X)     multimedia applications. In contrast to bitonal
                             --
         X SERVER                           X I E SERVER          imaging, these appl ications will requ ire the use of
                                                                  hardware accelerators wel l into the future.
                                  DEVICE DEPENDENT (DDX)
                                                                    Other appl ications wil l require richer user inter­
              l                                    +
         I             I
                                                                  faces util izing continuous-tone images, video, and


                                        I                   I
                                             IMAGING              aud io. All of these new data types are generally
             DISPLAY
                                             HARDWARE             data-intensive, and compression or decompression
                                                                  of any one of them is a significant processing bur­
                                                                  den. Hand l ing them in combination indicates that
             Figure 7       X/XIE Architecture                    the need for specialized hardware assistance will
                                                                  persist for the foreseeable future.

An  example of a routine at this level might be                   Continuous-tone Images
lmageDisplay, which d isplays an image in a previ­                Bi tonal images are either black or white at each
ously created window. lmageDisplay parameters                     point, but some appl ications requ ire smoothly
might include x and y scaling values, the rotation                shaded or colored images. These images are typi­
angle, and region-of-interest coord inates. Whether               cally referred to as continuous-tone images, a term
programming with the XIE protocol at the l ibrary                 that denotes e i ther color or gray-scale, e . g . , photo­
or toolk i t level, applications developers benefit               graphs, X rays, and still video. The representation
from the platform i nteroperability of the standard               and required processing of this image format is
interface. Image accelerator hardware and opti­                   significantly different from that of bitonal images.
mized device-dependent XI E code changes the                         Continuous-tone images are represented by mul­
application's image display performance, but an                   tiple bits per pixel. This format al lows a greater
application developed using the XIE protocol can                  range of values for each pixel, which yields greater
run on any XJE-capable server.                                    accuracy in the representation of the original
                                                                  object. Additionally, each pixel can consist of mul­
Accelerator Performance Results                                   tiple components, as i n the case of color. The num­
With the DECimage I200 X terminal, we have                        ber of bits used to represent a continuous-tone
achieved i nteractive performance rates, reduced                  image is chosen according to the nature of the
memory usage, and increased final image legibil i ty.             image.
We achieved these rates by transporting com­                          For example, medical X rays require a high
pressed files instead of huge pixel files and by imple­           degree of accuracy. Consequently, 12 bits are gener­
menting specialized image processing hardware.                    ally regarded as the minimum acceptable for the
The DECimage 1200 can read, transport, decom­                     rendering of this class of image. Color images typi­
press, scale, and d isplay an 8.5 -by-1 1-inch b itonal           cally require 8 bits per pixel for each component
document in I to 2 seconds. Successive displays,                  (YUV or RGB format) for a total of 24 bits per pixel.
i . e . , rotating, region-of-interest zooming, panning           Table 2 shows the relative size of samples of each
around the image, all occur in less than I second,                image. The need to express these images in a



D igital Techtzical ]ournal      Vol. 3 No. 4      Fall 1991                                                              21
Image Processing, Video Terminals, and Printer Technologies




compressed format is obvious from the storage                     because any change to either the evolving standard
space requi rements and the current storage med ia                or a standard extension coul d be easily introduced
limits.                                                           to the firmware. The fastest implementations arc
     The compression of cont inuous- tone images can              achieved by special-purpose hardware accelerators.
be accomplished in several ways. However, most                      The ]PEG i mplementation docs not requ ire hard­
imaging appl ications are            not closed        systems;   ware, i . e . , the algorithm can be performed com­
inevitably, each system needs to manipulate i mages               pletely in software. The case for hardware assist
that are not of i ts own making. for this reason we               is made in performance. Table 5 describes t he
acloptccl the ]PEG standard , which specifics an algo­            reduced instruction set computer (!USC) processor
rithm for the compression of gray-scale and color                 performance, in m i l l ions of operations per seconcl
images. Specifically, the ]PEG compression method                 (mops), needed to provide the specified operation
is based on the two-dimensional (20) d iscrete                    at a motion video rate of 30 frames per second 7
cosine transform (OCT). The OCT decomposes an                     However, generic IUSC processors of those speeds
8-by-H rectangle of p ixels into i ts 64 20 spatial­              are not avai lable today Therefore, dedicated, ctts­
frequency components. The sum of these 64 20                      tom very large-scale integration (VLSI) devices
sinusoids exactly reconstructs the 8-by-8 rectangle.              (such as the CL550-10 from C - Cube Microsystems)
However, the rectangle is approximated-and com­                   must be used to perform the operations 8 Even
pression is achieved-by discarding most of the                    if the motion video rate is not requi red , the ASIC
64 components. Typically adjacent pixel values                    devices offer the simplest hardware solution.
vary slowly, thus there is l i ttle energy in most of the
d iscarded high-frequency components.
                                                                  Live Video and Video Compression
     The edges of objects generally contribute to the
                                                                  Video captures the natural progression of events in
high-frequency components of an i mage, whereas
                                                                  an environment, and is the refore a natu ra l and
the low-frequency components are made up of
                                                                  efficient way to comm unicate. Consider, for exam­
intensities that vary more grad ually The more
                                                                  ple, the assembly of a set of components. One way
frequency components included in the approxi­
                                                                  to express the assembly process is to show a series
mation, the more accurate the approx i ma tion
                                                                  of photographs of the assembly at successive steps
becomes. Table 4 shows some sample ]PEG image
                                                                  of completion. As an alterna tive, video can show
compression ratios -"
                                                                  the actual assembly process from start to finish.
     The most popular part of the ]PEG stanclarcl ,
                                                                  Subtle details of the process such as part rotations
the "' baseline" method , was defined t o be easily
                                                                  and movements can be clearly conveyed , with the
mapped into software, firmware, or hardware.
                                                                  added dimension of time.
Straightforward OCT algorithms can be efficiently
                                                                    Obviously, information expressed in video form
implemented in firmware for program mable DSP
                                                                  can be valuable; however, significant problems arise
chips, clue to their pipel ined architecture. The fi rst
                                                                  in adap t i ng video for use in computer systems.
systems to embody the standard did so using DSPs,
                                                                  First, the huge data size of video appl ications can
                                                                  strain the system's storage capabi l i ty. Video can
Ta ble 4         Ty pica l Compression Parameters                 be characterized as a stream of continuous-tone
                 for J P EG                                       images. Each of these images consists of p ixel val­
                                                                  ues with individual components making up each
Compression           Com pression    Rendered I mage             p ixel . For video to have fu l l effectiveness, the sti l l
Ratio                 Method          I ntegrity
                                                                  i mages must b e presented at video rates. I n many
          2:1         Lossless        H i g hest qual ity-        ctses the rate to faithfu l ly reproduce motion is
                                      no data loss                30 frames per second, wh ich means that one
       1 2:1          Lossy           Excellent quality-          minute of u ncompressed v ideo (512-by-480 p ixels
                                      i n d isti n g u i shable   a t 24 bits per pixel) would consume over 1 gigabyte
                                      from the original
                                                                  of storage. In addi tion to storage demands, large
       32:1           Lossy           Good qual ity-              vo lu mes   of data      cause     bandwidth     problems.
                                      satisfactory for
                                                                  Presenting 30 frames per second to the video out­
                                      most appl ications
                                                                  put with the above parameters wou ld require a
      1 00 : 1        Lossy           Low qual ity-
                                                                  transfer rate of more than 22 megabytes per sec­
                                      recognizable
                                                                  ond from the storage device to the video outpu t.



22                                                                        Vol. 3 No. 4   Fall l991   D igital Techuicaljournal
                                                                         Hardware Acceleratorsfor Bitonal /mage Processing



Table 5       Processing Requ i re m e nts for I ma g i n g Functions

                                                                                                                 Processor
I ma g i ng                                     Processor Operations per Pixel*                                  Operations
Functions                     Read              Write             ALUt           M u ltiply         Tota l       at 30 fps (mops)

Pixel move                          .25            .25              0                 0                   .5              15

Point operation                 2                                    1                0               4                  1 20

3 X 3 convolve                  9                                   8                 9              27                  81 0

8 X 8 DCT                      24                                  14               16               65                 1 950

8 x 8 block                  1 28                                 1 91                0            320                  9600
match ing

'RISC processor, 1M pixels, 30 frames per second (Ips), 8 bits.
tALU = arithmetic logic unit




Thus, reducing the amount of data used to repre­                           Audio and A udio Compression
sent the video stream would al leviate both storage                        Video is usual ly accompanied by audio. The audio
and bandwidth concerns.                                                    can be reproduced as i t was recorded (with the
   The starting point for the compression of video                         video), or it can be mixed with the video from
is with still i mages and, as previously mentioned,                        a separate source (such as a compact elise (CD)
the J PEG algorithm can be used to compress sti l l                        player). The audio data is defined by appl ication
continuous-tone i mages. Because video can b e rep­                        requirements. I f the application a l l ows lower
resented as a sequence of still images, the algori th m                    qua l ity, the audio can be sampled at lower rates
coul d b e appl ied to each sti l l . T h i s procedure                    with fewer bits per sample, such as telephony rates,
wou ld produce a sequence of compressed video                              which are sampled at 8 kilohertz and 8 b i ts per
frames, each frame independent of the other                                sample. For appl ications requ iring high-quality
frames in the sequence.                                                    (CD) audio, samples are usu a lly taken at 44 k ilo­
   The evolving Motion Picture Experts Group                               hertz and 16 b i ts per sample.
(MPEG) standard takes advantage of frame-to-frame                            Integrating a u d io data into an application creates
similarities in a video sequence, thereby enabling                         special problems. The major characteristic that
more efficient compression than the appl ication                           d ifferenti a tes audio from the other data formats
of the ]PEG algorithm alone.9 I n most situations,                         presented here is i ts continuous nature. Audio
video sequences contain high degrees of s i m ilar­                        must flow u ni n terrupted for it to convey any mean­
ity between adjacent frames. The compression of                            ing. In video systems, the flow of frames may slow
video can be increased by encoding a frame using                           down u nder heavy system loading. The user may
only the differences from the previous frame. The                          never notice i t , or may not be annoyed by it. Audio,
majority of scenes can be greatly compressed; how­                         however, cannot slow or stop. For this reason, large
ever, scene transitions, l ighting changes, or condi­                      buffers are u sed to al low for load variations that
t ions of extreme motion need to be compressed as                          may affect a u d io reproduction.
independent frames.                                                          A more subtle problem i n creating applications
  The need for hardware assist in this area is com­                        using audio is in synchronization. Audio data is
pel l ing. Table 5 shows that to sustain a .JPEG decom­                    usually included to add another dimension of infor­
pression at 30 frames per second wou l d require a                         mation    to   the   app lication   (such   as speech).
1950-mops processor. The same result can be                                Without a method of synchronizing the video and
obtained using the CL550-10 ]PEG Image Compres­                            audio, one data stream w i l l drift out of phase with
sion Processor.9 Although this device does not                             the other. One way to include synchronization is to
make use of interframe similarities to increase com­                       use time stamps o n the audio and video. This is par­
pression efficiency, a device implementing the                             ticularly useful because standard time codes are
M PEG standard would exploit these similarities.                           used in most production machines.
Table 5 shows that motion compensation, to be                                The compression of audio data is not as efficient
supported at 30 frames per second, requires a                              as that of the other data formats. Since a statistical
9600-mops processor.                                                       approach to coding audio is h ighly dependent on



Digital Technical journal        Vol. 3 No. 4   Fall 1991                                                                       23
I mage Processing, Video Terminals, and Printer Technologies




the type of input (i.e. , voice, musical instru ment),             referenced. As the data is i ncorporated , it is com­
another method i s required for generalized inpu ts.               p ressed and stored. Authors requ i re the capabi l ity
Differential pulse code modulation (DPC .VI) is often              to edi t and mix video and audio passages to get t he
used to encode aud io data. DPCM codes only the dif                desired resu lt. Moreover, the video and audio may
ference between ad j acent sample values. Since the                originate from d ifferent devices and may. even be in
difference i n value between samples is usua l l y less            d ifferent formats.
than the magnitude of the sample, modest compres­                    As defined above, " user systems" do not require
sion can be achieved (4 : 1) . The l im i tation using this        all of the functions that aut horing systems need:
technique is i n the codi ng of h igh-frequency data.              only. decompression is requ i red i n a typical user
     Hardware assist for the audio data format w i l l             system. Most ex isting user systems require a n ana­
probably come i n t h e for m of hardware t o perform              log video source (videodisk), which is purchased
functions other than compressi o n . For instance,                 as part of the appl ication. The device control is per­
DSP algorithms can perform equa l ization, noise                   formed by the application, i .e . , when a user selects
reduction, ami special effects.                                    a passage to be replayed, the application sends
                                                                   commands to the videodisk. Figure 8 depicts an
Multimedia                                                         authoring system and a user system, along with
As the term impl ies, m u l t imedia may integrate a l l           suggested I/O capabi l i ty.
o f the previously. mentioned image formats. The                     Next-generation m u ltimedia platforms w i l l make
word " may." is important in this context. This area               fu l l use of d igital video and audio. This impl ies that
has been mainly technology-driven, due to such                     systems wil l be able to receive and transm i t m u lti­
factors as l ack of standards, developing 1/0 devices,             media appl ications and data over networks. This
insufficient system bandwidth, differing clara for­                interactive capabil i t y will improve the efficiency of
m ats, and a vast amount of software integra tion.                 many m u ndane appl ications and devices. For exam­
     It is currently a topic of debate whether typical             ple, electronic mail can be extended with video and
users will require the abi l i ty to create. as opposed            audio annotations, or meetings can be transformed
to only access, m u lt i media source material. How­               into v ideo teleconferencing. The adoption of com­
ever, for d iscussion p u rposes, m u ltimedia plat­               pletely d igital data for multimedia a lso implies that
forms     can   be   classified    into       two   categories:    the platform I/O w i l l change. Some user systems
authoring and user. Authoring refers to creation                   wil I not requ i re ana log device i nterfaces or contro l :
of multimedia source m a terial and req u i res differ­            t h e u s e r will load the application over the network
ent capabi li ties than user p latforms. In the creation           or from a n optical disk.
of a m u ltimedia application, data from many differ­                Each of the i mage formats described i n this
ent devices may need to be d igi tized and cross-                  section has different characteristics, and each wil l



                         AUTHORING                                                                   USER




                      110 A N D D E V I C E
                                                                                              110 A N D D E V I C E
                      CONTROL
                                                                                              CONTROL
                      COMPRESSION/
                                                                                              DECOMPRESSION
                      DECOMPRESSION




                          NETWORK                                                                 NETWORK




                                          Figure 8      Sample Multimedia Platforms




24                                                                         Vol. .3   To. 'I   Fall 1991    Digital Technical journal
                                                           Hat"dware Acceleratorsfor Bitonal image Processing



be presented in the embodiment of mul ti media.              References and Note
Given the size, processing requirements (compres­
                                                             1 . Standardization of Group 3 Facsimile Appa­
sion and decompression), and real-time demands of
                                                                 ratus for Document Transmission, cenT
applications, hardware assist w i l l be a necessity.
                                                                 Recommendations, Vol u me VI-Fascicle VII.3,
                                                                 Recommendation T.4 (1980).
Summary
Imaging is a un ique data type with special sys­             2. Facsimile Coding Schemes and Coding Control
tem requ irements. To achieve interactive rates of              Functions for Group 4 Facsimile Apparatus,
bitonal image d isplay performance today, hardware              CCITI Recommendations, Volume Vl l - Fascicle
accelerators are needed ; that has been the primary             VIJ .3, Recommendat ion T.6 (1984).
focus of this paper. In the future, a general-purpose
                                                             3. Digital Compression and Coding of Continuous­
processor should be able to hand le the imaging pro­
                                                                Tone Still Images, Part I, Requirements and
cess at the necessary speed, and beyond that, the
                                                                Guidelines, ISO!IEC JTC 1 Draft International
processor shou ld be affordable in a low-cost b i tonal
                                                                Standard 10918-1 (November 1991 ).
imaging system. However, the bitonal document
processing market wil l not wait; it is in a high state      4. J. Mauro, X Image Extension Concepts, Version
of growth and requ ires that products l i ke accelera­          2. 4 (Cambridge: MIT X Consortium, June 1988).
tors be developed for at least a few years.
                                                             5. D. A. Huffman, "A Method for the Construction
   Continuous- tone docu ments and mu ltimedia
                                                                of Minimum Redundancy Codes," Proceedings
applica t ions will place an even heavier processing
                                                                IRE, vol. 40 (1962): 1098- 1 101 .
load on an imaging system. These areas wil l require
accelerators for several years. As imaging applica­          6. G. K. Wa l lace, "The .JPEG Still Picture Compres­
tions, including bitonal, expand to cover more mar­             sion Algorithm," Communications of the ACIVI,
kets, the quali ty enhancements and performance                 vol. 34, no. 4 (April 1991 ): 30-44.
benchmarks met by accelerators today wil l set
                                                             7.   Table 5 is adapted from Y. Kim, " Image Com­
customer expectations. Consequently, our future
                                                                  puting Requirements for the 1990's: From Multi­
imaging products must be designed to meet these
                                                                  media to Medicine," Proceedings of Electronic
expectations.
                                                                  Imaging West (April 1991 ).

Acknowledgments                                              8. CL550 ]PEG Image Compression Processor, Pre­
The authors wish to express thanks to the                       liminary Data Book (San Jose, CA: C- Cube
X Window Terminal Hardware and Software Design                  Microsystems Inc., November 1990).
Groups for their support in developing the                   9. Coding of Moving Pictures and Associated
DECimage 1200 option . The two major ASICs used                 Audio, Com mittee Draft of Standard ISO 1 1 172:
in the design were developed for previous pro­                  ISO/MPEG 90/ 176 (December 1990).
jects, and those two design teams are also offered
our thanks. Special thanks to Frank G lazer and
Tim Hel lman for their insightful research on the
image rendering process.




Digital Tech-nical journal   Vol. 3 No. 4   f'i:lll 1991                                                       25
                                                                                                           Bjorn Eng berg
                                                                                                        T homas Porcher




X Window Terminals

                X window terminals occupy a nicbe between X window workstations and grapbics
                terminals. Tbe pU!pose of terminals in general is to provide low-cost user access to
                bost computers or smaller dedicated systems. X window terminals furtber tbe
                advance in grapbics terminals and prm•ide new and interesting wczys to utilize bost
                systems. Etbernet cable provides for grapbics performance previous�}' not seen in
                terminals. Tbe X Window 5)!stem developed by MIT allows multiple applications
                to be displayed and contml/ed from tbe user's workstation. Now, with X window
                terminals, tbe same poweJful user inte1face is available on bost and other non­
                workstation computers.



In mid 1987, the Video, I mage and Print Systems            graphics, but their performance was far below the
(VIPS) Group began the design of Digital's first            expectations of a workstation user. Few p eople
X window terminal, the VTIOOO terminal and its              have the patience to ru n , for example, a compu ter­
code upgrade, the VT 1200 terminal. Our goal was to         aided design appl ication on a VT240 terminal, assum­
design and implement an X window terminal that              ing such a version of the application is available.
would a l l ow the use of windowing capabi l ities on          Although a workstation offers fast graphics capa­
large computer systems. I n 1989, Digi tal developed        b i l ities, its applica tions sometimes need more CPU
the VT l ) OO X terminal and in 1991 the VXT 2000           power or more disk space to do calculations in a
X terminal. The designs of these X window termi­            timely fash ion. Graphics appl ications written for
nals are all quite different. Our design approach           workstations cou ld not run on faster host comput­
changed as the u nderlying technology changed.              ers, which did not provide a d isplay. Nor was there
     This paper first compares host -system comput­         a standard way to get data from the host to d isplay
ing with appl ications that run on workstations.            on a workstation. Each appl ication requ i red a
It summarizes the significance of the X Window              unique solution to this problem.
System developed by MIT and discusses the cl ient­             Since the introduction o f the new c l ient-server
server model. The paper then present s the need for         model of compu ting and modern networks, many
X window terminals and fol lows their development           tasks can be d ivided into subtasks that can r u n
stages. It compares and contrasts Digi tal's differ­        on t h e most suitable processor. The X Window Sys­
ent design strategies for the VTlOOO, VT I200, and          tem uses the cl ient-server approach, as shown in
VT I300 X terminals. The paper concl uues with a            Figure 1 . The appl ication is viewed as an X c l ient,
sum mary of the recently annou nced VXT 2000                and a workstation or a terminal can run an X server
X termina l .                                               that controls the display. The X server also controls
                                                            inp u t from the keyboard and mouse or other point­
Background                                                  ing devices.
Before the development of the X Window Sys­
tem, there was very l i ttle overlap in fu nctional i ty
between workstations and other kinds of compu t­
ers. Workstations had stunning and fast graphics,
and many powerful applications were avai lable on
them . Those applications were not available to users
of basic 80 -by-24 character-cel l text d isplay termi­
nals connected to a host system located in a clean
room. Graphics t<.:rminals, of course, al lowed the use
of ReGIS or another protocol for m a t h and busin<.:ss                Figure 1        Client-seruer Model




26                                                                 Vr>l. 3 No. 4   /-(i/1 /'}')/   D igital Teclmical journal
                                                                                                       X Window Terminals




  An X cl ient and an X server use an X wire to                        CPU as the X server, or it may execute on a host,
com municate, as shown in Figure 2. The X wire                         another workstation, or a compute server. The
is simply a two -way error-free byte stream , which                    X server can be connected to several X cl ients
can be i mplemented in many different ways. The                        simultaneously, with any combination of local
X Window System archi tecture does not stipu­                          (running on the same CPU) or remote (running on
late how the X wire shou ld be implemented , b u t                     another CPU) X clients. The X server treats local and
several d e facto standards have emerged . Manu­                       remote clients equally.
facturers have designed X wires usually based on
the clata transport mechanisms that were avai lable                    Workstation Environment
and convenient when the X Window System was                            Figure 3 compares a traditional non-X windowing
implemented. The X wires use transmission control                      workstation with an X windowing workstation. In
protocol/internet protocol (TCP/IP), DECnet, Local                     both workstations the appl ication must use a
Area Transport (LAT), and o t her protocols, and even                  graphics l ibrary to com mu nicate with the d isplay
shared memory buffers as a transport to avoid                          hardware and software.
protocol overhead. A single implementation often                          In an X windowing cl ient environment, the
supports several transport mechanisms.                                 l ibrary of routines is cal led Xl i b . An application
  The X server typica l ly executes on a processor                     designer can choose from a wide variety of toolk its ,
with display hardware. The X cl ient can execute on                    which are essentially a level of addi tional l ibrary
almost any processor. It may execute on the same                       routines between the application and Xlib. The use
                                                                       of a tool kit can significantly reduce the amount
                                                                       of work an appl ication programmer has to do. The
                                                                       appl ication software, Xl ib, optional too l kit, and
                                                                       other l ibraries compose the X cl ient, as shown in
                                                                       Figure 4.
                                                                          Wi th few exceptions, the X server comes with
                                                                       the d isplay hardware and input devices (keyboard
                                                                       and pointer) indicated in Figure 5.
                                                                          The X Window System with its flexibil i ty neatly
                                                                       solves the problems of CPU power and d isk space
                                                                       versus d isplay availabi l ity. Applications written for
                                                                       X can execute on a wide variety of compu ters, and
                                                                       the resu lts can be d isplayed on any of a m u l t i tude
                 Figure 2       X Wires                                of devices, even on a workstation that would not


                        TRADITIONAL                                    X W I N DOWING
                        WORKSTATION                                    WORKSTATION



                                APPLICATION     I                               A P PLICATION


                                GRAPH ICS
                                L I B RARY




                                                  I
                                WINDOWING
                                SOFTWARE

                                DISPLAY HARDWARE


                                I                         I
                            KEYBOARD    I     I       SCREEN




                                    I   MOUSE
                                                      I
                                             Figure 3          Inside the Workstation




Digital Teclmicnl Journal     Vol. 3 No. 4   Fall 1991                                                                       27
Image Processing, Video Terminals, and Printer Technologies




                                                                                                     stations. In a software development or document
          APPLICATION
                                                                                                     edi ting environment, the users often set up their
     OPTIONAL TOOL·KIT                                             X CLIENT                          workstat ·ions as terminals. They usually created a
     AND OTHER LIBRARIES
                                                                                                     few termina1l emu lation windows and used SET
              XLIB
                                                                                                      HOST or RLOGIN commands to connect to a host
                                                                                                     system on which they stored their working envi­
                                                                                                     ronment and fiks. Only two features of a work­

                   Figure 4     1he X Client                                                         station were frequently used . Users kept several
                                                                                                     terminal emulators on their screens a t the same
                                                                                                     time, and set the terminal emulator windows to be
                                                                                                     larger than 80 by 24 characters. Only rare.ly d id t he
     X PROTOCOL H A N D L E R
                                                                  X SERVER                           average workstation user take advantage of the fu l l
       DIS PLAY AND
       I N P U T HARDWARE                                                                            power o f graphics applications.
                                                                                                           The resu lts of our study ind icated a need for
                                                                                                     a cost-effective a l ternative to a workstation that
                   Figure 5     The X Server                                                         wou ld provide the features desi red by a large num­
                                                                                                     ber of users. We envisioned a new k ind of termi­
have the capacity to run the app l ication local ly.                                                 nal, one that would al low people to have m u ltiple
Figure 6 shows how the X Window System fits into                                                     windows of arbitrary size, to connect with m u l­
a network environment.                                                                               tiple hosts, ami, since the X architecture al lowed it,
     The X Window System has already generated                                                       to be able to use the same kind of graphics as a
many useful appl ications, and its widespread popu­                                                  workstation.
larity ensures that many more applica tions will be                                                        From an X arch i tecture standpoint, X terminals
made available in the future.                                                                        and X workstations are qui te similar. They can in
                                                                                                     fact use the same hardware. For example, Digital's
Needfor X Terminals                                                                                  VT1300 term inal runs on the same hardware as the
In a study to determine how worksta tions are used,                                                  VA.'Cstation 3100 worksta tion. X terminal software
the VIPS Group fou nd that many users did not take                                                   can also be made to run wel l on hardware plat­
advantage of the fu ll potential of their work-                                                      forms that are not sui table for worksta tions.




                                HOST CPU 1                                 HOST C P U 2




                                     I    X CLIENT       I                         1x       C L I E NT    1 1x   CLIENT   1 r       x CLIENTl
                                               I
                                                                                              l                   l                   1
                                 I       NETWORK
                                         I N T E RFACE:
                                                             I                     I                NETWORK I N T E R FACE
                                                                                                                                               I
                                               I                                                                   I                      ETH E R N ET



                                                                     I    X W I N DOWING WORKSTATION



                                                     I       NETWORK
                                                             I N T E R FACE:
                                                                               I
                                                                                       I   X C L I E NT   I I    X CLIENT l
                                                                                                I                   I
                                                      I                                    X SERVER
                                                                                                                            I
                                                                                                I                  l
                                                     I       KEY B OAR D       II           MOUSE
                                                                                                          II     SCREEN         I

                                              Figure 6               X Window Network Environment




28                                                                                                               Vol. .3 Nu. 4       Fat/ 1991     Digital Techrtica l ]ournal
                                                                                                    X Window Terminals




  The   main    architectural           d ifference   between        host, via the Ethernet or the serial l ines as shown
the X terminal and X workstation software is that                    in Figure 7. This capabi l i ty m a kes the VT1200 ter­
X terminals are closed systems that do not sup­                      minal useful in an environment that does not have
port local user applications. Although this may                      X window support.
seem to be an unnecessary restriction, i t does al low                 Although any X server can ru n windows soft­
X terminals to be made for less money. An open sys­                  ware, it does not provide a user interface. To manip­
tem that a l l ows any user appl ication to run local l y            u late the windows, the user needs a window
m u s t have a n established CPU architecture, a sup­                manager. The window manager creates window
ported operating system, such as the VMS, UNIX,                      frames that al low the user to invoke fu nctions to
or ULTRIX system, and, subsequently, sufficient                      move windows, resize windows, change stacking
memory and/or disk space to support such an envi­                    order, and use icons. This capabil i ty also makes the
ronment. A closed system, on the other hand, can                     VT 1200 terminal useful when no host is available to
be designed with simpler hardware, a smaller oper­                   run a remote window manager. A terminal with a
ating system, less memory, and thus lower cost.                      local window manager generates less network
The absence of the abili t y to run user applications                traffic, and window management i s not slowed
locally does not i mpact usability significantly since               by host congestion or network round-trip delays.
the user can run any desired application on another                  The VT1200 X terminal allows use of a remote win­
CPU. D igital's VTlOOO and VT1200 X terminals were                   dow manager, if the user prefers a different style of
designed based on this approach.                                     window management.
                                                                       The local terminal manager provides the user
X Te rminal Environment                                              i nterface to initiate connections to host systems.
X terminals often have local applicatio ns, but they                 It is also responsible for the terminal customization
must be built i n to the terminal by the designers.                  interface.
The VT1200 terminal has a video terminal emulator                      All clients commu nicate with the X server using
(VTE), a window manager, and a terminal m anager                     standard X wire comm ands only. Any window man­
as the local applications. The VTE a ll ows the VT1200               ager, remote or local, can manage all the windows
terminal to m ake American National Standards                        on the screen, regard less of whether the clients are
Institute (ANSI) character-cel l connections to a                    remote or local.



                             HOST C P U 1             HOST CPU 2




                            S E R IAL
                            LINES




                                           Figure 7        The X Terminal Environment




Digital Technical journal     Vol. 3 No. 4     Fall 1991                                                                 29
Image Processing, Video Te rminals, and Printer Tech nologies




Development of X Window Terminals                            col without some compression of the wire protocol
The development process of the VT 1000 and                   i tself. We had to build D igi tal's first X terminal with
VT1 200 X terminals has important lessons to teach           an Ethernet interface.
us. The knowledge we gained in 1987 has helped us                We needed to determine if this hardware platform
develop future generations of X terminals.                   could give us sufficient performance. We made sev·
   When we designed the VTlOOO X terminal and                 eral performance estimates, based on what we
its code upgrade, the VTl200, we held many discus­           knew then about the X server and other software
sions within the group and with people from other             components. We went through each step in as much
groups. We planned many i terations before we                detail as we could (before anything was built) . We
arrived at the final architecture. It was by no means        calcula ted how many instructions were necessary
the only way to design an X terminal, and in 19H9            to perform each task in the chain of receiving a
we tried a different approach with the design of the         command and displaying i t on the screen. By .know­
VT1300 terminal . We knew that the best decision at          ing the speed of the CPU, we could estimate per·
a particu lar time might be very d ifferent from the         formane<.: in characters or vectors per second .
best decision one year later, since the technical and        Our estima tes showed that the VTIOOO X terminal
marketing environment is constantly changing.                would not be exceedingly fast, but the perfor·
New tools, standards, and practices enter the field          mance would most probably be sufficient, de.fi·
while others become obsolete. Newer products                 ni tely faster than a VA.'I(station 2000 in most cases.
must always have new features to meet changing                   In retrospect, actual performance of the VT1000
technology requirements.                                     terminal and the later software upgrade , the
                                                             VT1200, was close to our estimates, but i t took sev­
                                                             eral passes of code optimization to ach ieve such
Hardware Platform
                                                             performance .
 Our .first step was to discuss the hardware plat­              We also discussed alternate hardware designs
 fo rm and select the kind of CPU to use, memory             for p erformance improvements. One solution pro·
size, 110 considerations, type of d isplay, etc. We          posed two CPUs, the TMS34010 microprocessor to
stud ied many diffe rent CPUs to determine which
                                                             hand le the display and a 68000 microprocessor to
one would provide the most capabi l ities for the            hand le I/0 and other tasks. Unfortunately, we fou nd
lowest cost. A VAX chip was rejected because, at the         no easy way to balance the workload between the
 time, it was far too expensive for the required price       two crus. We estimated that the different software
 range of the VTIOOO terminal. The Motorola 68000            components would have the fo llowing relative CPU
series CPUs are quite powerful, but we had to con·           demands:
sicler other factors such as ava i labi l i ty of software
and hardware tools, cross comp ilers and l inkers            •   Interrupts, 5 percent
that could run on the VMS system, and hardware
                                                             •   Com munications, lO percent
debugging facil ities of sufficient power. We fma l ly
selected Texas Instruments' TMS34010 micro·                  •   Operating system, 5 percent
processor with video support and several built -in
graphics instructions that made it a cost-effec tive         •   X server (minus d isplay routines), 60 percent
sol ution. It also came with VMS development tools,          •   Display routines, 20 percent
a c compiler, an assembler and linker, a single-step,
hardware trace buffer with disassembler, and a                  To equalize the load between the CPUs, we would
powerful in-circ u i t emulator that made it possible        have had to spli t the X server in two, a solution that
to control execution in detail, inspect registers and        was not feasible. Any other spl i t of tasks wou ld
memory, and set break poi nts and hardware watch             cause one CPU to spend most of i ts time waiting
points (for example, break when writing value .x             for the other, and the overal l performance gain
into locationy).                                             would be minimal. Communication between m u l·
    We further discussed the kind of l/0 to use. A           tiple CPCs is complex and is very d ifficult to debug.
sample implementation of the MIT X server on a               Therefore, we decided that two CPUs were not
VAXstation 2000 workstation and a primi tive serial          worth the trouble or the cost. The best way to
l ine protocol showed, as expected , that serial l i nes     double performance is to insta l l a single CPU that
were clearly insufficient to carry the X wire proto·         is twice as fast. At that time, the TMS34020 was




30                                                                   Vol. _) No. 4   Fall 19')1   D igital Teclmical]ou>·naf
                                                                                              X Window Terminals




al ready being mentioned as a fol low-u p m icro­             u ncovered l ittle timing windows and race condi­
processor. Since its software wou ld be compatible            tions that had not been hand led properly. Even
wi th the TMS34010, we decided to keep it in mind             w i th i n-circu it e m u l a tors, such bugs could take
for possible use in a future terminal.                        weeks to track clown.
                                                                In the VT1300 we decided to use the VAXELN
Code Selection                                                operating system. We wanted to avoid the possibil­
                                                              ity of time wasted on finding and patching holes in
The use of read-only memory (ROM)-based code
                                                              the design of a new operating system.
versus downloaded code has been debated for
some time. ROM-based code starts up faster and
incurs less network traffic at startup time (espe­            Local Terminal Manager
cia l ly on a site with m a n y X terminals), but is not      The VTlOOO X terminal is self-starting at power-up,
flexible when software is upgraded. On the other              b u t without a host system, it needs a local user
hand, downloaded code can be easily distributed.              i n terface. We decided that this i nterface shou ld
An entire site can be upgraded with one or a few              resemble a workstation session manager and thus
installations by a system manager as opposed to               called it the local terminal manager. Although it
changing ROMs in a large number of terminals.                 covers a different set of functions, we wanted the
(Wit h the VT1200 X terminal, customers can change            local terminal manager to implement a similar set of
ROM boards.) From the point of view of terminal               objects and operations (the "look and feel" or style)
business, it made sense to use ROM-based code in              of a workstation session m anager. The style of the
1987. We reasoned that not a l l sites would have             DECwindows session manager was chosen to make
E thernet, but with ROMs the X terminal would                 it easier for a user to swi tc h between an X terminal
still be usefu l as a m u ltiwindow terminal emu la­          and a DECwindows workstation. We wrote a subset
tor. We real ized that such concerns would change             toolk i t for all the " customize" screens and ensured
with time, and on t he whole, downloaded code                 that the VTE could use the same subset tool k i t for
would become the better approach. The only                    its "customize" screens. As DECwindows has pro­
exceptions wou ld be i n the home or small office             gressed, subsequent X terminals have adapted the
markets where a boot host or an Ethernet might not            new user interface preferences, i n this case Motif.
be avai lable. Subsequent X termina ls are being
made in both downloaded (for example, in the                  Local Terminal Emulator
VT1300 terminal) and ROM versions.                            We considered a local terminal emu la tor to be an
                                                              important component. We knew that X-based ter­
Operating System Selection                                    minal emulators could run on the host, but in 1987
Next we considered which operating system to                  hosts with X windowing support were rare. Since
use. We looked at other vendors' operating sys­               we were in the terminal group, a terminal that
tems, but found they were either too complex and              could not manip u l a te ordinary text by i tself was
big or inadequ a te . One of our coworkers had writ­          considered u nsel lable . We wanted the ability to
teo a very compact operating system for a VAX                 access both X and non-X hosts and we wanted
system used on a nother project. We used it in our            to support m u ltiple text windows. Therefore we
prototype and then adapted i t for the TMS34010               defined the terminal emu lator as an X cl ient so that
processor. We implemented add i t ional functions             text windows could coexist with X client windows.
to run the rest of the software with minimum                  This feature has proved to be exception a l ly popu­
changes.                                                      lar. A large n u m ber of users use nothing but v ideo
  There are many advantages to working with                   terminal emulator windows. They are not inter­
"your own" operating system. It is easy to make               ested in X windowing graphics, but do want mul­
changes, to work around trick)' problems, and to              tiple and/or larger text windows on a l arge screen.
make special enhancements. But operating system
code is difficult to debug. Timing is very critical,          Local Window Manager
and throughout the project, we fou nd strange bugs            We debated whether or not to implement a local
in code that had i n i t ia l ly appeared to be a l l right   window manager. The DECwindows window man­
to everyone involved . We fou nd bugs u nder heavy            ager was u nder development and was constantly
load condi tions after a rare sequence of events              changing.   The DECwindows window manager




Digital Technical journal    Vol. 3 No. 4   Fall 1991                                                              31
 Image Processing, Video Termina ls, and Printer Technologies




contained far too many VMS depende ncies to be              Communications Protocol
ported easily. Also the X terminal did not have             Many com mu nications protocols were available,
enough memory to run the DECwindows wol k i t               but our choice was d ic ta ted by market pressures
local l y. We cou l d have ported other window man­         rather than technical reasons. The market demanded
agers, but they lacked the essential characteristics        TCP/IP. DECnet would have been acceptable, but
of the DECwindows window manager. For a wh i le             it was running out of avai l able addresses, a t least
we considered letting the local clients have a primi­       within D igital. DECnet address space supports
tive way £0 manage their own windows, until a ful l ­       only 64,000 nodes and requires manual address
featured window manager could be started o n a              and name assign ments. After waiting weeks to get
host. Again, this alternative lacked the DI:Cwindows        addresses for a few worksta tions, we real i zed that
system's quali ties. We eventu a l ly decided to write      adding thousands of X terminals into Digi ta l 's inter­
a window manager based only on X l i b and our              nal network wou ld not be possible. DECnet Phase v
subset toolk i t calls. It has the essential characteris­   software has solved this problem.
tics of the DECwindows product. Al so, s ince the             Next we looked at the LAT protocol used by
DECwindows window manager of necessity wou ld               Digital terminal servers and fou nd that i t bad sev­
keep changing, we wrote the l ocal window man­              eral advantages. first, the VMS operating system
ager in such a way that it cou ld rel i nquish control      supports the LAT protocol. LAT uses u nique 48 -bit
to a remote window manager. This solution gave us           Ethernet addresses to iden tify each node, which
the most flexibility for this hardware platform. The        al lows a large node address space. LAT a lso does not
recently announced VXT 2000 X terminal has been             requ ire any system management to add another ter­
designed with virtual memory to accom modate a              minal. A user can connect a terminal to a power
well-establ is heel unmodified window manager, the          source, and the terminal a u tomatically becomes
Motif Window Manager.                                       part of the network. Our performance evaluati ons
                                                            found that the LAT interface on the host could be
X Server                                                    writ ten to incu r Jess host overhead than DECnet,
                                                            which is important when many X terminals are con­
We a l s o needed to choose an X se rver. We cou ld
                                                            nected to hosts.
have based our code on the d istribut ion tape from
                                                              Changes were needed i n the VMS LAT d river to
M IT, bu t a t the time the X Window System was not
                                                            accommodate X wire and font service connections.
yet a mature product. Every implementor had to
                                                            The VMS Software Engineering Group worked with
spend considerable time stabilizing the implemen­
                                                            us to ensure that we would have those changes
ta tion enough to yield a prod uct and i mprove per­
                                                            on schedule and in the appropriate VMS releases.
formance. Since the VMS DECwindows Group had
                                                            As a resu lt, we chose the LAT protocol for the VMS
been writing code for the server, we decided to use
                                                            com munity and TCP/IP for users of ULTRIX and UNIX
DECwindows code. Once the porting effor t started,
                                                            systems.
we found that most of the performance had been
improved by V;'w'{ MACRO cod e. Consequen tly, we
had to re-engineer a l l the modu les or adapt new          Font File System
ones from the MIT tape. As we kept porting and              Storing fonts and changing fo nt file formats were
enhancing performance, our code changed more                major problems. Since the VT IOOO X terminal d id
and more until it became extremely difficult to             not have a local file system, some fonts had to be
track bug fixes made by the DECwindows Group.               stored in ROM to a l low the VT lOOO terminal to func­
The MIT patches were also nearly impossible to use          tion in standalone mode. A quick review of the
because of code changes and because our starti ng           available DECwinJows fon ts showed that not a l l of
code was one step removed from the tape.                    them fit in the ROM space al lowed for the term inal .
     1b<Jay the MIT X server is a mature product;           Furthermore, customer-designed fon ts or new font
pa tches and bug fixes are readily available from           releases could not be accom modated. The solution
MIT and from the X com m u n i t y. In our current          was to be able to read fo nts from a host system.
X terminals, the h igh degree of portability of t he        This approach provided a fo n t service on the VMS
M IT X server al lows us to keep most of the :\-! IT        system, and enabled fo n t files to be read over the
X server source code almost unchanged so patches            Internet. We designed a process cal led the fon t dae­
are easily appl ied.                                        mon to run on the VMS operating system. This pro-




32                                                                 Vol. .:i No 4   Fa ii i'J'JI   Digital Techuicaljounwl
                                                                                           X Window Terminals




cess cou ld del iver font data on request to one or        nate format, the VT1200 terminal does not have to
several VTIOOO terminals. The VMS system 's fon t          open and read the font file until a cl ient actually
daemon uses the LAT protocol t o del iver t h e fonts      i n tends to use it.
and protects somewhat against font file form a t
changes. In many ways, t h e design o f the font           Comparison of X Terminals
daemon makes it a precursor to a general fo nt
                                                           The VT1200 and VT1300 X window terminals
server, and it is very similar to the X Font Server
                                                           were built using different approaches to solve
being delivered by MIT in the latest release of the
                                                           the problems encountered du ring development.
X Window System.
                                                           The X terminal is a new and flex ible concept; there
    To use the font service, the terminal user must
                                                           is no single "best" design. Table 1 compares the
specify a font path in the VT1200 local terminal
                                                           most important differences between the two termi­
manager. Specifying a host name is sufficient to
                                                           nals. We also i nclude the specifics for the VXT 2000
access the default font path, al though users with
                                                           X terminal.
their own font files can option a l ly search other
                                                                The VT1200 is ROM-based; all its software is per­
directories. At startup, the VT1200 terminal makes a
                                                           manently resident in the terminal. The VT1300 soft­
font connection to the host's font service and del iv­
                                                           ware is downloaded, so a host or bootserver on the
ers the font path sp ecification to the font service.
                                                           same network must supply the terminal w i th a load
The fo nt service sends font names and other basic
                                                           image at power-up.
font information about a l l the fonts in the selected
                                                                Since downloaded terminals are dependent on
path. When the VT1200 X server needs a font, the
                                                           the existence of at least one working host system,
VT1200 first searches the ROM-based fo nts; if it is
                                                           the user interface can be designed differently.
not there, a request to read the fo nt is sent to the
                                                           While the VT1200 X terminal has a built-in user
font daemon. The daemon sends the requ i red infor­
                                                           interface, the VT1300 does not need it. The VT1300
mation to the VT1200, and the X server can display
                                                           terminal automatical ly makes an X connection to a
characters from that font. Since memory is l i mited,
                                                           host at power-up, and the user is presented with
the VT1200 has fon t caching, a mechanism to dis­
                                                           the same DECwindows login box as on a work­
card fonts no longer used or to d iscard the least
                                                           station. The VT1300 has no local cl ients; a l l clients
used fonts. Our current X terminals increase the
                                                           run on the host system.
 robustness of the font mechanism ; for example,
                                                                The VT1200 terminal uses the LAT protocol for
they provide recovery should the font service or i ts
                                                            i ts ease of use and minimal network management
host become unavailable.
                                                           demands. The VT1300 terminal uses the DECnet
    The special LAT code that we used on VMS sys­
                                                           software al ready implemented in the VAXELN oper­
 tems for the font service was not available on
                                                           ating system used internal ly. Both terminals sup­
 UNIX and U LTRJX operating systems. Si nce inter­
                                                            port TCP/IP.
 net protocol (IP) was available, we could use the
 trivial file transfer protocol (TFTP) to read a file
from a host system, if the system manager set the          VXT 2000 X Terminal
 proper protections. We chose TFTP for its ease            One problem that has plagued all X terminals is
 of implementation and its wide availabili ty on           limited memory space. Workstations usual ly have a
UNIX and U LTRJX systems. The TFTP font path i n a         virtual memory system, which provides large pag­
 VT1200 terminal specifies a host I P address and a        ing and swap areas on a disk, and applications
 complete path to a file (usually named font.paths)        and X servers can use more memory space than
 that contains the complete path to a l l the font         the hardware has. Until now X terminals have not
 files that the VT1200 can use. The terminal can           had virtual memory systems. If too many appl ica­
 then access a l l those font files, again through TFTP,   tions made excessive demands, or if a cl ient created
 to obtain font names and other basic i nformation         large off-screen images (ca lled " pixmaps" in the
 about each font. When a client wishes to use a fon t,     X Window System) the terminals quickly used all
 the proper font file can be read again, this time to      memory space . If the X server implementation
 load the complete font. Since this process is time­       was correct, an error was reported and a client
 consuming, the font path pointing to the file has         might try a less demanding approach. I n other
 an alternate format in which the font name fol­           cases, the terminal or cl ient might simply crash.
 lows the complete path to each file. Using this alter-    One alternative was to install more memory in the



Digital Teclmicaljournal    \1()/. 3 No. 4   Fall 199l                                                          33
Image Processing, Video Te rminals, and Printer Tec hnologies




Ta ble 1 Compa rison of X Wi nd ow Te rminals

VT1200 Te rminal                              VT11300 Te rminal                             VXT 2000 Terminal

Monochrome only                               Color only                                     Monochrome and color
1 bit plane                                   4 or 8 bit planes                              1 o r 8 bit planes
Code i n ROM                                  Code downloaded                               Code downloaded
No virtual memory                             No virtual memory                             Virtual me mory
2-4 M B RAM                                   8-32MB RAM                                    4-1 6 M B RAM
TMS34010 CPU                                  VAX CPU                                       VAX C P U
Special operati n g system                    VAXELN operating system                       Special operating system
Local c l i ents:                             No local c l i e nts                           Local c l i e nts:
Te rminal manager                                                                           Te r m i nal manager
Win dow manager                                                                              Motif win dow manager
Video term inal e m u lator                                                                  DECterm term in al emu lator
Local customization                           Customized on host                            Local custo m i zation
                                              just as a workstation                         Central i zed customization

Choice of host (LAT on ly)                    Automatic X w i n d ow                        Choice of host
                                              log in to boot host                           (LAT and TCP/ I P using XDMCP)

LAT protocol                                  DEC n et protocol                             LAT protocol
TCP/1 P protocol                              TCP/I P protocol                              TCP/I P protocol

Special hardware                              Avai lable on several                         Uses standard hardware
                                              workstation platforms




X terminal, although this can be costly and offers no                customization of a l l X termin;�ls on the network.
guarantees.                                                          Figure 8 shows the configuration for the VXT 2000
     In the next generation of Dig i t a l 's X term inals,          X terminal.
the VXT 2000, this problem has found a cost­
effective solution. Based on the VA.,'( arc h i tecture,             Conclusion
t he VXT 2000 term inal uses virtual memory and                      X term inals are not in tended to replace work­
downloaded code.          The Digital I n.foScrver,        an        St:ltions. Nor wi l l workstations repl ace host sys­
Ethernet storage se rver, provides the load image,                   tems or completely d isplace X terminals in the
virtual memory paging space, fonts, and customiza­                   foreseeable fut ure. It is l i kely that host computers
tion storage. The same I nfoServer al so solves                      w i l l a lways be faster and h ave m ore memory and
another problem: now the X terminal has access to                    disk space than reasonably priced worksta tions
a file system. This a l lows more extensive customi­                 of the same era. It is a lso l ikely that term inals can
zation, as wel l as cen tral ized management of the                  be built cheaper than workstations of reasonable




                                   Figure 8     The VXT 2000 Network Environment




34                                                                          Vol. j No. 4   Fa/1 1991   Digital Technical journal
                                                       X Window Terminals




performance for some time to come. As long as that
is the case, there will be a market for X terminals
and host systems. Future X terminals will be faster,
and have more built-in functionality, more local
appl ications, X extensions, and most l ikely, addi­
tional hardware fea tures. X terminals will be the
networked terminals of the 1990s.

Acknowledgments
We wish to thank the members of the VT1200 devel­
opment team who worked many long hours on this
project. Thanks to everyone inside and outside
the Video, Image and Print Systems Group who
contributed helpful suggestions, constructive criti­
cism, and important hours using and testing the
products. Thanks to the LAT and VMS Software
Engineering Groups for incorporating the changes
needed for the VT1200 X terminal to be useful.
Thanks to the VIPS Quali ty Group for ensuring that
as few bugs as possible remained in the product
when shipped.




Digital Tech11icaljournal   Vol. 3 No. 4   Fall /991                  35
                                                                                                         Peter A. Sichel I




ACCESS. bus, an Open Desktop Bus
                With the recent introduction of the ACCESS. bus product, Digital has affirmed its
                commitment to open systems and thus to facilitating better solutions for inter­
                active compu ting. This open desktop bus protJides a simple, uniform way to link
                a desktop computer to as many as 14 low-speed 1/0 devices such as a keyboard,
                mouse, tablet, or three-dimensional tracker ACCESS. bus features a 100-kilobit-per­
                second ma.:'<imum data rate, hardware arbitration, dynamic reconjiguration, a
                mature capabilities grammar to support generic device drivers, and off- the-shelf,
                low-cost flC microcontroller technology



As the cost of personal interactive com pu ting             describes the process we went through to convert
decreases, the range of appl ications and the need          to a new bus arch i tecture, and summarizes the key
for specialized I/0 devices is growing dramatica l l y.     advantages of the chosen design.
Traditional personal compu ters were designed to              The basic desktOp bus concept is i l lustrated i n
accept only a small number of standard devices;             Figure 1 . The b u s a l lows m u ltiple, low-speed l/0
adding devices beyond those origin a l ly envisioned        devices to be interconnected and thus i nterfaced
usually requ ires special ized hardware or software.        through a single host port. Desktop bus devices
Custom i nterfacing is expensive for vendors and            such as a keyboard or a tablet, which are not hand­
users and thus li mits the availability of new devices.     held, provide two connectors and al low another
     ACCESS. bus provides a simple, u n iform way tO        device to be da isychained. A hand-held device
l i n k a desktop computer to a nu mber of low-speed        such as a mouse can be placed at the end of the
l/0 devices such as a keyboard, a mouse, a tablet, or       daisychai n , or a con nector expansion box can be
a   three-d imensional (3-D) tracker. Designed from         a t tached to accommodate add i tional devices that
the begin n i ng as an open desktop bus, ACCESS . bus       do not provide two connectors.
fac i l i tates cooperative solutions using equ ipment
from different vendors. This paper describes the
ACCESS. bus design and gives some insight intO how
the idea was adopted at Digit a l .

Design Goal, Process, and Advantages
The design goal for the desktop bus fol lows from
our experience within the Video, Image and Print
Systems (VIPS) Input Device Group wi th trying to
support new devices on Digi tal terminals and
workstations. While various new devices have been
                                                                             CONN ECTOR
successfu l ly prototyped over the years, the need                           EXPANSION
for nonstandard hardware and custom software                                 BOX
drivers was always an expensive, time-consumi ng
obstacle. Even after successful prototyping, these
devices could not be read ily adapted tO our stan­
dard systems, l i m i ti ng their use to custom appl ica­
tions. In designing the desktop bus, our goal was to
make i t as easy as possible to i nterface previously
unavailable 110 devices to our systems in a way
that was both practical and marketable. This sec­
tion expla ins the benefits of u sing a des ktop bus,                   Figure 1        Basic Desktop Bus


36                                                                 Vol. 3 No. 'I   Fa/1 /'J'J/   Digital Technical journal
                                                                                    ACCESS.bus, an Open Desktop Bus



    The desktop bus has the following benefits:                   By m a k i ng the bus truly open, we encourage third
                                                                  parties to add value to our systems.
•   Enables greater flexibility and variety of use                  The benefits of a desktop bus are significant. But
•   Reduces the cost of connecting m u ltiple devices             converting to a new architecture, especial ly one
                                                                  that is not back-ward compatible, is expensive in
•   Expedites bringing new technology to market                   terms of the time and effort requ ired . How does a
                                                                  large corporation b u i ld agreement to make such
•   Helps leverage third-party devices
                                                                  an investment decision' The desktop bus project
    The first benefit, greater flexibil i ty, can be simply       started as a grass roots engineering effort and grad­
achieved by al lowing addi tional devices and more                ually bu ilt momentu m . The process was one of
modular solu tions. We further extended this bene­                dialogue to attract partners. Initial ly, three groups
fit by designing a way for devices to be added at run             with sl ightly different objectives worked together
time without disrupting system operation. Con­                    to develop the bus. The visibi l i ty of separate groups
figuration s hould be automatic; connecting stan­                 jointly supporting the bus concept was essential to
dard devices shou ld not requ ire powering down or                transform the idea into actio n . People are more
rebooting the system before a new device can be                   wil l ing to accept an idea that others around them
used. The desktop bus supports m u l t iple l ike                 have already adopted.
devices wi thout switches or j umpers.                              The three groups that i n i t i a ted the desktop
    The second benefit, reduced cost, was cruc i a l to           bus project were our VI PS Input Device Group i n
having the bus accepted as a solu tion across a wide              Westford, MA , mentioned prev iously; t h e Work­
range of products from low-end v ideo terminals                   station Systems Engineering (WSE) Grou p, located
to high-end workstations. We recognized that con­                 in Palo Alto, CA; and the Video Advanced Develop­
temporary electrical techniques could e liminate                  ment (A/D) Group in Al buquerque, Nl\1. Our Input
the need for level translation circ u i ts, - 12 vol t (V)        Device Group was looking for ways to simpl ify the
power sup pi ies, and p erhaps some of the protec­                process of prototyping special ized input devices
tive components used with RS-232 interfacing.                     and of getting related software support for our
Although many devices would now require two                       video terminals and worksta tions. WSE was devel­
connectors, system cost wou ld decrease because                   oping a low-cost, personal workstation and needed
we wou ld need to supply only as many connectors                  a flexible way to support m u l t iple input devices
as the number of devices to be attached , or possibly             without greatly increasing the cost of the base
one more.                                                         workstation. The Albuquerque A/D Group had been
    The third benefit, expedi ting the time to market             experimenting with next genera tion 1/0 devices,
for new technology, a l lows u s to better satisfy                i . e . , force-feedback joystick, 3-D tracker, and real­
changing requ irements. Key to this benefit is hav­               time audio and video, and was interested in having
ing the means to connect new devices without                      these technologies adopted by other Digital groups.
changing the system hardware or software. Based                   This AID Group had used PC technology success­
on our experience with input devices, we devel­                   fully in one of its previous video projects.
oped the concept of device capabi l i ty reporting                  In january of 1990, engineers from each group
and generic device protocols. Standard devices                    real ized they were working on similar problems
like keyboards and locators, e . g . , mice, tablets, and         and began to col laborate. The WSE Group was to
trackba lls, all work in similar ways. For t h is class           build the desktop bus host interface and software
of device, we define a simple device protocol and                 drivers into their worksta tion; the VlPS Group was
a way to parameterize and report device unique                    to help define the device protocols and supply
characteristics. A single generic driver can adapt                desktop bus keyboards and mice; and the Albu­
i tself to work with a class of similar devices so                querque AID Group was to support bus devel­
that no custom software is required for basic opera­              opment and prototype addi tional devices. Within
tion of standard devices.                                         four m o nths, VIPS had defined the basic protocols
    Leveraging     third-party    devices,     the       fourth   and could demonstrate a working FC keyboard
benefit, is a imed at satisfying d iverse customer                and mouse. These early prototypes helped per­
requ irements. Because the use of computers con­                  suade WSE to support the project and, in turn,
t inues to proliferate, t he range of appl ications far           helped reinforce the importance of the project to
exceeds that which any one vendor can master.                     the VIPS Group.




Digital Tecbnicaf jounral     Vol. 3 No. 4   Fall 1991                                                                  37
I mage Processing, Video Te rminals, and Printer Technologies




   We began presenting the desktop bus idea to                 •   Dynamic recontiguration. The hardware and
interested groups w i thin Digi tal and received many              software al low bus devices to be " hot -plugged "
useful suggestions includ ing                                      and used immedia tely, without restarting the
                                                                   system. The devices are recognized automa ti­
•    Use the same keycodes as on the LK201 keyboard
                                                                   cally and assigned un ique add resses. This advan­
     to el i m inate the need to rewrite k<:yboard
                                                                   tage results in a plug-and-play user interface.
     lookup tables.
                                                               •   A mature capabil ities grammar to support generic
•    Store the country keyboard variat ion inside
                                                                   device drivers. An extensible free-form grammar
     the keyboard so users wil l not need to enter i t
                                                                   allows devices to describe their characteristics
     manual ly.
                                                                   to a generic driver. Most common devices can
•    Keep the devi ces simple, without modes.                      work with standard drivers.

 In addi tion, third-party input device vendors                    Bus or network interconnection has become
made the fo.l lowing suggestions.                              widely accepted as a means of provid ing flexible
                                                               open solutions. To appreciate ACCESS. bus, it is help­
•    Usc a modular connector that is easy to plug and          fu l to position its performance capabili ties with
     unplug correctly.                                         respect to those of other network interconnect
•    Provide enough power for several additional               technologies, as shown in Table 1 .
     devices.

•    Al low vendors to supply their own device                 Ta ble 1     Netwo rk I nterconn ects
     drivers; tuning their own device drivers is part
                                                                                                      Order of Magnitude
     of the value added by the vendor.
                                                                                                      Performance
                                                               Bus Ty pe                              (kilobits per second)
   The bus idea was elegant and genera lly wel l
received . Most of the reservations centered around            Apple DeskTo p Bus,                    10-1 00
the l ikely impact on existing system components,              ACCESS.bus
the current problems, and whether conversion to                LocaiTalk                              1 00-1 ,000
the bus was feasible. Because we recognized that               Ethernet                               1 ,000-10,000
other groups were facing tight dcvclopmcnt schecl­             FDDI                                   1 0 ,00 0-1 00,000
ulcs, we did not pressure these groups to support
our desktop bus work. We presented the desktop
bus as a possible solu tion to i nterface problems,              At first glance, the 100-kb/s speed of the
made our design information available, and worked              ACCESS .bus may seem adequate for large desktop
to incorporate suggestions. But as the development             devices l ike printers and modems. Bu t these
work progressed, more partners supported our                   devices can transm it long data streams indepen­
effort.                                                        dent of any user activity and , if not restricted, could
   Once we cleciclecl to use a des ktop bus, we                compromise the interactive performance of the
looked at available designs, including the Apple               bus. Thus, ACCES S . bus is intended for low-speed
DeskTop Bus, the Musical Instrument Digital                    activities that people perform with their hands
Interface (MIDI), and serial buses offered by other            and is fast enough to hand le mul tiple interactive
semiconductor vendors, and evaluated these al ter­             devices l ike a keyboard, mouse, or 3-D tracker.
natives with respect to our design goal. Key advan­
tagcs of the design chosen, i.e., the ACCESS. bus, are         Hardware Description
                                                               Before discussing the ACCESS. bus design, we pre­
•    Off- the-shclf in terintegrated circu it ( J 2C) micro­
                                                               sent a description of the Philips FC technology
     controller technology with maximum data rate
                                                               upon which the design is based. Details of the
     of 100 kilobits per second (kb/s) . This technol­
                                                               specific ACCESS. bus implementation fol low.
     ogy is low-cost, yet fast enough for sophisticated
     input devices l ike a 3-D tracker.                        Jnterintegrated Circuit Fundamentals
•    Built-in hardware arbitra tion, which simpl ifies         ACCESS. bus extends t he Phil ips PC bus to operate
     the software and allows reliable com munication           off-board and , t hus, connect des ktop devices. The
     without inventing a new protocol.                         F C is a two-wire serial clock and serial data




38                                                                     Vol. 3 No. 1   Fall I'J'JI   Digital Technicaljmwnal
                                                                              ACCESS.bus, an Open Desktop Bus



open-collector bus. An open-collector design means               maximum capacitance not to exceed 700 pico­
that the clock and data l ines are normally in a high­           farads (pF).
impedance floating state and are pulled up to a log­
                                                             •   Number of devices. The m aximum number of
ical high state.
                                                                 ACCESS.bus devices allowed on the bus is 14.
   A device that wants to send a message waits for
                                                                 Limiting factors are the device addressing range
any message frame in progress to complete, then
                                                                 and the power distribution (a total of 500 rnA for
asserts a START signal to become bus master and
                                                                 all devices).
begins to generate data and clock signals. The bus
clock is synchronized among a l l devices by its             •   Hardware interfaces. ACCESS.bus hardware inter­
wired AND connection. Each device, whether                       faces are implemented using standard PC micro­
transmitting or receiving, stretches the low period              controllers developed by the Signetics Company
of the clock u ntil ready for the next bit to be trans­          or under l icense from Philips Corporation. (Sig­
ferred. When the last device is ready, the bus clock             netics Company is a d ivision of North American
is al lowed to go h igh, generating a rising edge on             Phil ips Corporation.)
the serial clock. At this time, all active devices
sense the state of the bus data I ine . For a receiving      ACCESS.bus Protocol
device, the state represents the received data bit.
                                                             Every device on the bus is a microcontroller with
For a transmitting device, the state determines
                                                             an FC interface and behaves as ei ther a master
whether the device has successfu l ly asserted its
                                                             transmitter or a slave receiver, exclusively, as
data on the bus. A transmitter that is sending a logi­
                                                             defined by the FC Bus Specification.
cal high state and detects that the data l ine is being
held low by another sender, recognizes that it has
                                                             Message Format
l ost arbitration and must try again later. When a
"co l l ision" or arbitration occurs, no data is lost, one   A message transmits information between a device
                                                             and the computer or between the computer and one
message is transmi t ted and received, and the
remaining messages must be sent again.                       or more devices. There is one exception: a device
    FC data messages are transmitted as 8-bit bytes,         may attempt to reset other devices assigned to the
with each byte being acknowledged by a ninth                 same address by sending a Reset message to i tself.
                                                               ACCESS . bus messages have the fol lowing format:
ACKNO\VLEDGE bit from the receiver. PC technol­
ogy also defines unique START and STOP signals to
                                                             Byte             Bit Number
delimit message frames. The first byte of any mes­
                                                             Number           1 2 3 4 5 6 7 8 l
sage frame is always the destination address.
                                                                                   destaddr     10 l   Destination
ACCESS. bus Physical Implementation                                                                    address
Details of the physical implementation of ACCESS. bus        2                     srcaddr      10 l   Source
are as fol lows:                                                                                       address
•   Basic electrical configuration. ACCESS. bus uses         3              [P I       length          Protocol
    four-pin, shielded, modular-type connectors that                                                   flag, length
    feature positive orientation and locking tabs.                                                     (the number
    Data and power for the bus are transmitted over                                                    of data bytes
    low-capacitance, four-wire, shielded cable. The                                                    from 0 to 127)
    fou r conductors are used for grou nd, serial data,
    serial clock, and + 12 v.                                4 through                body             Consists
                                                             (length + 3)                              of 0 to 127
•   Available power. The maximum available power                                                       data bytes
    for all devices is 12 v at 500 milli amperes (rn A).
    ACCESS. bus devices may supply their own power           length + 4            checksum
    from a separate source, if needed. A power-up
                                                               Initial ly, devices respond to a defa u lt power-up
    reset circu it must still be provided to reset the
                                                             address. During the configuration process, the com­
    device when bus power is appl ied.
                                                             puter assigns a u nique address to every device on
•   Cable length. The maximum cable length for               the bus. Messages are ei ther device data stream
    the entire bus is 8 meters. The limi ting factor is a    (P=O) or control/status ( P = l ) , as indicated by the



Digital Teclmicaljour11al   Vol. 3 No. 4   Fall 1991                                                                 39
Image Processing, Video Terminals, and Printer Technologies




protocol flag. The minimum length of a message is                    unique address, and connect devices to the appropri­
4 bytes; the maximum length is 131 bytes (127 data                   a te software driver. Configu ration norma'lly occurs
bytes and 4 bytes for overhead). The message                         at system start-up, or at any time when the com­
checksum is computed as the logical XOR of all prc­                  puter detects the addition or removal of a device.
vic)lls bytes, including the message address.
                                                                     Power-up/Reset Phase
Standard Messages                                                    When reset or powered up, a device a lways reverts
The ACCESS . bus protocol defines the seven stan­                    to the default address and sends an Attention
dard interface messages summarized in Table 2.                       message to alert the compu ter to its presence. At
Parameters defined within the body of the message                    system start-up or reini tialization, the compu ter
are listed in parentheses.                                           sends a Reset message to all PC addresses in the
                                                                     ACCESS. bus device address range (14 messages) to
Identification                                                       ensure that all devices on the bus respond at the
Since the ACCESS. bus is a bus-topology network,                     power-up default address.
unique identification strings arc used to distinguish
devices. These strings arc structured as fol lows:                   Identification Phase
protocol revision:        1 byte (e.g. , "A'')                      To begin address assignment, the computer sends
module revision:          7 bytes (e.g . ''X l . 3
                                         .               )          an Identification message at the device default
                                                     "


vendor name:              8 bytes (e .g. , " DEC             )      address. Every device at this address must then
                                                         "


module name:              8 bytes (e.g. , " LK501            )      respond with an Identification Reply message. As
                                                         "


device number:            32-bit signed integer                     each device sends its message, the JlC arbitra tion
                                                                    mechanism automatically separates the messages
   The module revision, vendor name, and module                     based on the identification strings. The computer
name strings an: left- justified ASCII character                    can then assign an address to each device by includ­
strings padded with spaces. The device number                       ing the matching identifica tion string in the Assign
string is a 32-bit two's complement signed integer                  Address message. When a device receives this mes­
and may be either a random number (if negative) or                  sage and finds a complete match with the identifi­
a unique serial number (if posi tive).                              cation string, it moves its device address to the new
                                                                    assigned value. As soon as a device has a unique
Configuration Process                                               address, it is allowed to send data to the computer.
The configuration process is used to detect what                      The FC physical bus protocol al lows mu ltiple
devices are present on the bus, assign each device a                devices on the bus at the same time if those devices


Ta ble 2    Sta n d a rd ACCESS.bus Protocol Messages

Com puter-to-device Messages                         Purpose

Reset ( )                                            Fo rce device to power-up state and defau lt 12C address.
Identification Request ( )                           Ask device for its "ide ntification st ring."
Assign Address (identification string,               Tel l device with matc h i n g "identification string" to change its
new add ress)                                        add ress to "new address."
Capab i l ities Request (offset)                     Ask device to send the fragment of its capab i l ities i nformation
                                                     that starts at "offset. "

Device-to-computer Messages

Attention (status)                                   I nform computer that a device has f i n ished i t s power-u p/reset
                                                     test and needs to be conf i g u red; "status" is the test resu lt.

Identification Reply (identification string)         Reply to Identification Request with device's un ique
                                                     "identification stri n g . "
Capabilities Reply (offset, data fragment)           R e p l y to Capab i l i t i es Req uest w i t h "data fragment," a fragment
                                                     of the device's capab i l ities st ring; the computer uses "offset"
                                                     to reassemble fragments.




40                                                                           v'<:!l. .3 No. 4   Fall /':)':)1   Digital Technical]ottrnal
                                                                                              ACCESS.bus, an Open Desktop Bus



are transmitting exactly the same message. In the                             puter then parses the capabil i ties string to choose
rare event that two l ike devices report the same                             the appropriate appl ication driver for the device.
random number or are mistakenly assigned to the                               The parsed string is also made available to applica­
same address, each interactive device transmits a                             tion programs using the device.
Reset message to its assigned address prior to send­
ing i ts first data message after being assigned a new                        Normal Operation
address. The self-addressed Reset message forces                              During normal operation, the computer periodi­
other devices at the same address back to the                                 cally requests inactive devices to identify them­
power-up default address, as if they had just been                            selves. If a device is found to be missing, or a new
hot-plugged . The message guarantees that each                                device appears by sending an Attention message a t
device has a unique address, but not until the                                the defau lt address, the computer sends an Identi­
device is actually used . The pseudo-random number                            fication Request message to each device address
(or serial number, if available) d istinguishes devices                       previously recorded as in use (up to 14 messages) to
at identification time before they are used, al lowing                        confirm which devices are still present. The com­
the host to inventory which devices are present.                              puter also sends a Reset message to each device
                                                                              address previously recorded as not in use. The com­
Capabilities Phase                                                            puter then begins the address assignment process
Device capabi lities is the set of information that                           by sending an Identification message to the default
describes the functional characteristics of an                                address and assigning each device that responds to
ACCESS. bus peripheral. The purpose of capabili ties                          an unused device address.
information is to al low software to recognize and
use the fea tures of bus devices without prior                                Gene ric Device Concepts
knowledge of their particular implementation. By                              ACCESS.bus uses the concept of generic device
having locator devices report their resolution, for                           drivers to support familiar 1/0 devices using only a
example, generic software can be written to sup­                              few drivers. Generic specifications for keyboards,
port a range of device resolutions. Capabil ities                             locators, and text devices have been developed.
information provides a level of device i ndepen­
dence and modularity.                                                         Keyboards
   The structure of capabilities information is
                                                                              The keyboard device protocol attempts to define
designed to be simple and compact for efficiency,
                                                                              the simplest set of functions from which a Digital
but also extensible to support new devices without
                                                                              LK201 or a common personal computer keyboard
requiring changes to existing software or periph­
                                                                              user interface can be bu ilt. A generic keyboard con­
erals. These objectives are supported by m aking
                                                                              sists of an array of key stations assigned numbers
the structure hierarchical and representing capabil­
                                                                              between 8 and 255. When any key station transi­
ities information in a form that appl ications (and
                                                                              tions between open and closed, the entire l ist of
humans) can use directly. The capabilities informa­
                                                                              key stations currently closed or depressed is trans­
tion is an ASCII string constructed from a simple,
                                                                              mitted to the host. This reporting scheme is func­
readable grammar. The grammar allows text strings
                                                                              tionally complete; the host can detect every key
to be formed into l ists, nested lists, and l ists with
                                                                              transition, and the scheme provides informa tion
tagged elements. The capabi l ities string for a loca­
                                                                              about the ful l state of the keyboard on each report.
tor might read as follows:
                                                                              No special resynchronization reports are requ i red .
( p ro t ( L o c a t o r )                                                       In addition to reporting key stations, the generic
    t ype ( mou s e )
                                                                              keyboard device can support simple feedback
    buttons(        1 ( L)   2 ( R )   3 ( M )   )
    d i m ( 2 )   rel   r e s ( 200    inch>         rang e ( -1 27   1 27)
                                                                              mechanisms such as keycl icks, bells, and l ight­
    d Q ( dname ( X ) )                                                       emi tting d iodes. These mechanisms are controlled
    d1 ( dname ( Y ) )
                                                                              explicitly from the host so that minimal keyboard
)
                                                                              state model ing is required. The capabili ties infor­
  After assigning a unique address to a device, the                           m a ti o n is used to identify the keyboard mapping
computer retrieves the device's capabi lities string                          table and the feedback mechanisms available. The
as a series of fragments using the Capabil ities                              keyboard mapping table can also be stored in the
Request and Capabil i ties Reply messages. The com-                           keyboard i tself as part of the capabi lities string.




Digital TechnicalJourna.l                Vol. 3 No. 4      Fa/11991                                                             4l
Image Processing, Video Terminals, and Printer Technologies




Locators                                                         appl ications. As the advantages of this open desk­
The locator device protocol is designed to accom­                top bus design become wel l known, we expect
motla te a range of basic locator devices such as                wider adoption of this product. The ACCESS. bus
a mouse or tablet. More complex devices can be                   is currently implemented on Digital's Personal
modeled as a combination of basic devices or can                 OECstation 5000 workstation, with implementa­
provide their own device driver, thus minimizing                 tions underway for the next generation of RISC
the burden on the protocol.                                      workstations and v ideo terminals.
   A generic locator consists of one or more dimen­
sions described by numeric values and , optional ly,             Acknowledgments
a small number of key swi rches. The standard driver             Many people contributed to the design ancl devel­
requires the locator device to identify the type of              opment of ACCESS.bus. I would especial ly l ike to
data it wil l report from a small list of options and            acknowledge Tom Stockcbrand and Tom Furlong
adjusts to hand le this data type. These options are             for their vision and early support; Chris Cued, Mark
                                                                 Shepard, and Ernie Soul iere for their contributions
•    Number of d imensions, e . g . , two, for a mouse or
                                                                 to the ACCESS. bus electrical design and protocols;
     a tablet
                                                                 and Robert Clemens for the excellent demonstra­
•    Dimension type: absol ute, i . e . , referenced to          tion hardware and firmware development support.
     some fixed origin, li ke a tablet: or relative. i . e . ,
     changed since last report, l i ke a mouse                   Gene ral Rej(e rences
•    Resolution in divisions per unit, e.g. , counts per         D. Lieberman, " Desktop Bus Is Born Free," Elec­
     inch or counts per revolut ion                              tronic Engineering Tintes (September 2, 1991 ) : 16.
•    Dynamic range of val ues that can be reported,              ACCESS. bus Developer's Kit (Palo Alto, CA: Digital
     i.e., the minimum and maximum values                        Equipment Corporation, Workstation                    Systems
•    Number of key switches, from 0 to 15                        Engineering TRJ/ADD Program, 1991 ).

   The assignment of scalar-value dimensions                     Signetics f l C Bus Specification (Sunnyvale, CA:
returned from one or more devices to the user                    Signetics Company, a Division of North American
interface functions is left to the appl ication. How­            Philips Corporation, February 1987).
ever, to accommodate most conventions. the scalar
d imensions and the key switches can be l abeled in
the capabilities string.

Text Devices
The text device protocol is intended to provide a
simple way to transmit character data to and from
character devices such as a bar code reader or a
sma l l character display. A generic text device trans­
mits a stream of 8-bit bytes from a character set.
Simple control messages are defined to support
flow control and to select communication parame­
t<.:rs that might be used to i nterface with a modem.
The capabilities string contains informa tion that
identifies the specific character set and communi­
cation parameters used.

Summary
The ACC.ESS. bus network interconnect offers the
possibility of a standardized, low-speed, plug-and­
play serial communications channel that can untan­
gle peripheral i nterfacing and open the way to new




42                                                                     Vol. ."> No. 4   t�{t/1 J')')J   Digital Teclmicaljournal
                                                                                               Ri chard Landau
                                                                                                 Alan Guenther




Design of the DECprint
Common Printer Supervisor
for VMS Systems

              DECprint Printing Services software controls a variety of printerfeaturesfor a wide
               range of printers. It supports several different page description languages, handles
               multiple media simultaneously, and uses different I/0 interconnections and commu­
               nication protocols. Operating within the VMS printing environment, it imple­
               ments a large number of user-specified options to the PRINT command. DECprint
               Printing Services functions as the supervisor in the VMS printing system for all
              PostScript printers supplied by Digital. The common printer supervisor bas an espe­
               ciallyflexible internal structure and processing method to serve complex printing
               environments.



The increasing variety and complexity of printing           the design of CPS, focusing on its capabilities within
devices in the last decade have strained the abi l i­       the VMS system. The paper then d iscusses the oper­
ties of operating systems to support them. Users            a tion of the VMS printing system and the enhanced
demand access to, and control over, the increas­            printing environment made possible by CPS.
ingly sophisticated features of their p rinters. At the
same time, appl ication programming resources are           Printing System Dirnensions
stretched by the requirement to support various             A printing system is the set of software and hard­
devices and features. Modern operating systems              ware components through which print requests
include printing systems that support printers and          pass from the time the user decides to print a docu­
insu late applications from many details of printing.       ment u ntil the appropriate hard copy arrives.
    DECprint Printing Services software was designed           The variety of printing devices in use is a chal­
to handle a wide variety of printers, with a range          lenge for the printing system and for applica­
of I/O connections, media hand l ing capabili ties,         tion program mers. We use the word "printer" in
fi n ishing equipment, data syntaxes, and so forth.         this article to imply the fu l l range of output devices
It provides the controll ing software that supports         that are at tached to systems and networks. A sys­
the fu ll range of Digital printers capable of printing     tem today must support a wide number of d imen­
PostScript documents.                                       sions: marking technologies, media, medium sizes,
    DECprint Printing Services functions as a compo­        speeds, transmission rates, and interconnects.
nent of the VMS printing system at the level of
printer supervisor, cal led symbiont in VMS termi­          The DECprint Model of Printing
nology. The supervisor is known within Digital as
                                                            The DECprint model of printing is composed of sev­
the DECprint common printer supervisor or com­
                                                            eral layers. Each layer has defined functions and 1/0
mon print symbiont (CPS). It is called common
                                                            interfaces. The layers of the DEC print model and their
because it replaces a number of different symbionts
                                                            relationships to VMS and CPS are shown in Figure 1 .
and is com mon to a range of printers. CPS is a com­
                                                            This model of print ing describes a usefu l structure
pletely new program developed by the Video,
                                                            with consistent functions and responsibilities.
Image and Print Systems Group.
    This paper explores the environment in which            •   Appl ication. An application program creates
printing systems now reside. It describes the struc­            information that the user may want to print. Al l
ture and functions of DECprint Printing Services and            types of applications fit into the model at this



Digital Techu'icaljourual   Vol. 3 No. 4   Fall /991                                                             43
Image Process ing, Video Te rminals, and Printer Technologies




                             DECPRINT ARCHITECTU R E                                  VMS                             CPS


                                              APPLICATION



                                USER                                              D IGITAL
           JOB                  INTERFACE                                         COMMAND

                            - - -1- - -                                          - - +- -
           SUBMISSION                                                             LANGUAGE
           INTERFACE                                       -   -




                                             PRINT
          ( N ETWORK)                        CLIENT                               SYS$SNDJBC

                                               1                                 - -+- -
          PRINTING
          INTERFACE         -   -   -   -     -    -   -   -   -




                                            PRINT                                  QUEUE
                                            SPOOLER                                MANAG E R
                                                               PR I NT
                                               I               SE AVICE                 I
                                    P R I NT                                       PRINT                           COMMON
          PRINTER                                                                                                  PRINT

                                - - - -1- - - -                                  - -+- -
                                    SUPERVISOR                                     SYMBIONT
          ACCESS                                                                                                   SYMBIONT
          INTE RFACE


                                            MARKING                                PRINTERS
                                            ENGINE

                                               I               PR INTER

                                        FINISHING
                                        EQUIPMENT



               Figure 1   Relationships of the Vll1S Printing System Components to the DECprint Model


     level, from data processi ng programs and simple                     •   Pri nt c l ient. The client accepts requests through
     text edi tors to high-quality document formatting                        its API, performs defau lting for the user, assists in
     and publ ishing appl ica tions The appl ica tion may
                                        .                                     selecting the correct prin t service, ga thers the
     pres e n t a printing interface directly to the user,                    print instructions and docu ment files, and sub­
     or may create a final form document from which                           mits the job to the print service. The protocol
     the user can access other print ing i nterfaces.                         used to submit the job may be operating system­
                                                                              specific or may be based on emerging standards
•    User printing interface. A user expresses the
                                                                              for network printing. The print service may be
     desire to prin t through a user interface to the
                                                                              local to the print cl ient (and the user), or it may
     prin ting system. The i nterface may be oriented
                                                                              be located elsewhere in t he network.
     to written com mands, to user selection of
     menu choices, or to a point -and-sdcct graph ical                    •   Print service. The print service is a convenient
     int e rfac e .                                                           abstraction that includes t he prin t spooler and aU
                                                                              su bsequent layers in the execution of the print
•   ]ob subm ission interface. User interface pro­
                                                                              job, for some set of physical printers. Printers
     grams com m unicate with the lower kv<.: b of the
                                                                              are often grou ped together based on their static
     printing system through an applic a tion program­
                                                                              characteristics, such as type of printer, printer
     ming interface (API) to the print client. The API
                                                                              data syntax, ancl defa u l t media.
     contains fu l l ca pabili ties for creating, d estroying         ,




     a nd man aging print jobs of all types. The job sub­                 •   Print spooler. The print spooler accepts the print
     mission i nterface may be operating svstcm­                              job from the client, spools the files and queues
     specific or may he based on emerging standards                           the job for later execution if necessary, ancl
     for network printing.                                                    then schedu les the job for execution. If the job



44                                                                                 Vol. 3 No. 4   Fall 1991   Digital Technical jou rnal
                                           Design of the DECprint Common Printer Supervisorfor VMS Systems



    requires resources that are not immediately avail­         Implementations of the model in various operat­
    able, human intervention may be necessary. For          ing systems and printers may express the layers
    example, if a job requires a special print medium,      differently, sometimes skipping certain layers. The
    then an operator or other printer attendant must        VMS printing system contains components at most
    provide the medium for the printer. If the job          levels of the DECprint model. The DECprint com­
    requires a special font, the spooler may be able to     mon printer supervisor (CPS) operates within the
    obtain the fon t from a l ibrary without human          VMS system, as indicated in Figure 1 . We designed
    intervention.                                           CPS to satisfy the requ irements and projected needs
                                                            of users, system managers, and programmers. In the
•   Printer supervisor. The supervisor d irectly con­
                                                            next section we discuss the design of CPS.
    trols the printer. It interprets the print i nstruc­
    tions for the job, manages the printer and its fin­     Sharing Devices
    ishing equipment, and writes the document data
                                                            Printers are often shared, especially high-end or
    to the page description language (PDL) inter­
                                                            specialized, expensive devices. Since shared print­
    preter. It also monitors the status of the printer,
                                                            ers are not always immediately available to the
    supplies some resources on demand, and responds
                                                            user or application program, the printing system
    to error conditions. On the VMS operating sys­
                                                            is required to hold jobs for printing later. The sys­
    tem, the printer supervisor is cal led a symbiont;
                                                            tem must be able to store the user's instructions for
    on ULTRIX and UNIX systems, a daemon.
                                                            printing, along with the contents of the document,
•   PDL interpreter. Generally, final form document         until they are needed.
    data is written in a data syntax intended for print­
    ing, but it is not in the native form required          Insulating the Application from Details
    by the marking engine. A PDL interpreter trans­         A printing system insulates appl ications from the
    forms the printer language into the lower-level         details of printing devices. For example, DECprint
    form for the marking engine. For example, in a typ­     Printing Services provides communications mecha­
    ical laser printer, a PostScript interpreter trans­     nisms and protocols, determines whether a shared
    forms the PostScript language into a device-level       device is currently busy, and sometimes translates
    bi t map and media control instructions for             printer data syntax.
    the print engine. In a simpler impact printer,              Appl ication programmers generally prefer to
    the controller turns characters and control             deal with as few external interfaces as needed to
    sequences into pin timing and paper movement            perform the task. Thus it is desirable to minimize
    instructions.                                           the number of d ifferent classes of printing devices
                                                            while maximizing the variety and flexibility of
•   Marking engine. The marking engine consists of
                                                            printing devices. The DECprint architecture speci­
    the media transport and printing mechanisms,
                                                            fies that the printing system take responsi b i l i ty for
    generally controlled at a low level. Marking may
                                                            matching the needs of the appl ication to the capa­
    be done by a wide spectrum of technologies, and
                                                            bilities of the output device, whenever possible.
    the media used may also vary widely. For the
                                                            For example, a printing system might need the abil­
    most part, descriptions in this paper use raster
                                                            i ty to transform the printer data stream from a
    devices such as laser printers as examples.
                                                            data syntax used by the application to a data syntax
•   Finishing equipment. The overall printing sys­          used by the printer. Hidden transformation makes
    tem includes finishing options that are not often       the system easier for applications to use. DECprint
    considered part of the (largely electronic) print­      Printing Services provides a certain number of
    ing system. Currently affordable components of          printer data syntax transformations of this type,
    the printing system are typically automated. For        from languages such as DEC PPL3 (which is com­
    example, several years ago duplex (two-sided)           monly referred to as "ANSI" within Digi tal) and
    printing was not economical for most office             ReGIS to PostScript, and from PostScript to printer
    applications; today it is, and many office printers     bit maps.
    include this finishing fea tu re. Stapli ng, on the
    other hand, is still not economical for most office     Internal Structure of CPS
    applications, though i t is i mplemented i n many       In designing CPS, our primary goal was to create a
    high-end production printers.                           flexible system that would hand le all the printer



Digital Technical journal   Vol. 3 No. 4   Fall 1991                                                              45
Image Processing, Video Te rminals, and Printer Technologies




features we could foresee ancll many tha t we could                as simple as "create a separator page·· or as complex
not foresee, a system that could be modified as                    as the sequence "read a file, perform carriage con­
needed to hand le not just new printers but new                    trol convt:rsion , add /HEADER, translate from A:'-ISI
classes of printers. CPS is capable of managing a                  data syntax to PostScript, a nd write the ITs u l t to
wide variety of character, l i ne , page, and document             the printer." A print job is a set of job steps that per­
printers.                                                          forms a l l fu nctions the user requests explici tly or
     lb create a flexible printing system, we needed to            implicitly. The CPS fac i l i t y that transla tes sekcted
design a highly modu lar internal structure. This inter­           prin ter data syntaxes into the PostScript language is
nal structure combines modules i n to sequences a t                discussed in the section Data Syntax Tra nslation.
several levels t o provide a general framework for
control l i ng and manipula ting I/O devices.
                                                                   Printjob A nalyzers
     At the bottom level of the structu re are fi lter
                                                                   To simpl ify the addition of new printers and new
modu les, which are l ightweight, independently
                                                                   cl asses of printers, CPS contains a software struc­
schedulable subprocesses within a VMS process.
Filter modu les commlmicate with each other by                     ture that corresponds to the hardware mechanisms
                                                                   of a printer.
means of 1/0 rou tines and a shared data structure
containing joh information. Pointers to the 110 rou­                 A print job ana lyzer (PJA) determines which

tines and shared data are suppl ied in the invoca­                 job steps are requ ired to process a job. CPS i ncludes
                                                                   a separate print job analyzer for each major class
tion of the filter module. The effect of the stream
                                                                   of printer that        it supports:     serial   PostScript,
1/0 routines is much l ike that of pipes i n the UNIX
                                                                   PrintServer, and LN03 I mage printer devices. When
operating systems.
                                                                   the symbiont begins execu tion, a PJA is chosen based
     A t the next h igher level is a set of com municating
                                                                   on the type of device associa ted with the queue.
filter modu les; each stream of filter modu les is
                                                                   This PJA is used u n t i l the symbiont is stopped. If a
cal led a job step. Final ly, a module cal led the print
                                                                   term inal device, such as a TI or TX or LT device, is
job ana lyzer combines a sequence of job steps to
                                                                   associated with the queu e , then the PJA for a serial
hand le a complete print job.
                                                                   device is i nvoked. I f a n LD device is used, then the
Filter Modules andjob Steps                                        PJA for an LN03Q printer is c hosen . Otherwise, the
Filter modu les can read input from a preceding fil ter            PJA associated with PrintServer devices is used.
modu le and write data to a succeeding filter mod­                   Each PJA contains a l ist of al l job steps requ i red to
u le. h lter mod u les may p erform fu nctions such as             execute a job on the class of pri n ters it supports.
reading a file, converting carriage control , translat­            The PJA selects the job steps it needs from this l ist,
ing data syntax, or writing data to the printer. A                 depending upon the i nstructions received from the
filter modu le receives as argu ments an input stream              queue manager.
and an ou tpu t stream , like a UNfX process, and a                  Job steps are I i nked together. The first job step
shared data structure, u n l ike a Ul\TJX process. A sim­          chosen by the PJA is l inked to the termination of the
ple fi l ter moclulc reads data from the inpu t stream,            P.JA i tself; when t he PJA finishes compil ing the job,
processes data, and wri tes data to the output stream.             it termina tes, thus starting the execution of the job.
  A fi l ter modu le may condition i ts operation based            At the begi n n i ng of each job step, each filter mod­
on informa tion from the shared data structure or                  u le is assigned stack space and a stack fra me. Its ini­
the contents of the data strea m. For example, a                   t i a l program counter address and arguments are
translator filter modu le might form a t data based on             stored in i ts saved registers for process activation.
the page size, margins, and aspect ratio specified                   CPS uses a piped stream 1/0 mechanism similar in
in the shared data structure, or based on control                  fu nction to a U N IX stream; a filter module 's input
sequences in the data stream, or bot h .                           comes from the outpu t of the previous module, and
     Not a l l filter modu les use t h e input o r output          its output becomes input to the following mod u le.
streams. The file reader filter module reads from the              I3y convention, the first filter mod u le of the job step
file instead of the input stream. Similarly, the device            is activated first i n the job step; when a filter blocks
output mod ule writes to the printer instead of the                for output, t he next fi lter module is activated . That
output stream .                                                    fil ter modu le then runs u n ti l i t blocks for inpu t or
     A j o b step i s a s e t o f filter modu les piped together   output, a t which point the previous or fol lowing
to perform one complete subtask. A subt:tsk may be                 filter module is activated .




46                                                                        Vr>l. 3 Nu. 4   Fall 1991   D igital Technical journal
                                               Design of the DECprint Common Printer Supervisorfor VMS Systems



Ta ble 1       Simpl ified Job-step Sequence                                 Table 1 shows a simplified listing of the job steps
                                                                          compiled by the serial PJA to process a simple job:
Job Step               Functi o n
                                                                          one file to be printed in ANSI mode. Each of the job
init_ps_device         Ensure the device is "fixed up."                   steps shown contains one or more filter modu les
check_prologues        Ensure that persistent                             piped together. For example the job-burst job step
                       prologues are loaded.                              has two modules piped together: the job-burst mod­
sheet_count            Get the beg i n n i ng page count.                 u le and the write-to-printer module. Figure 2 shows
job_burst              Print job bu rst page.                             several job steps with several filter modules each.
sheet_size             Get the cu rre nt sheet_size.
                                                                             If an error occurs at any point in the processing
                                                                          of a job, CPS skips job steps until it reaches the
wait_sheet_size        Wait for the sheet_size before
                       conti n u i ng .                                   identified error job step set by the PJA. In Table 1 ,
fi le_set up           Send a n y f i l e /SETUP mod u l es.
                                                                          the error job step points to the sync job step that
                                                                          precedes the job-trailer job step . In this case, CPS
get_vmbytes            Get the amount of local printer
                       memory avai lable on the                           resynchronizes with the printer and prints the job­
                       printer.                                           trailer page, includ ing the error message.
wait_vmbytes           Wait for the local printer
                       memory message f rom t h e                         Event Handling
                       pri nter.                                          In addition to the output side of processing a job,
file_out               Read the f i le to print and send                  there is a corresponding input side. The input side
                       it to the DECansi translator.                      reads messages from the printer, parses them , and
sync                   Wait for the printer to finish all                 notifies the appropriate hand ler of the event. The
                       pages.
                                                                          handler is chosen based on the type of message sent.
init_ps_device         Ensure the device is "fixed up."
sheet_count            Get the ending page count.
                                                                          •   CPS internal messages are d ispatched to the
                                                                              appropriate symbiont routines. For instance,
wait_sheet_count       Wait for the page count to
                       come back.                                             printer resource messages contain information
job_trai ler           Print the job trailer page.
                                                                              that affect CPS internal operations: paper size is
                                                                              stored for later use by layup (the general map­
sync                   Wait f o r the printer to finish
                       the job-trailer page.                                  ping of page images to sheets) and translators;
d i sconnect           Release the printer.
                                                                              virtual memory size is stored for translators; and
                                                                              page count is stored for la ter use in accou nting.




                      JOB STEPS

                      FILTER
                      MODULES




                         J                              READ GET
                                                        VMBYTES
                                                        MODULE
                                                                                WAIT
                                                                                FOR
                                                                                VMBYTES




                      Note that data flows from top to bottom and job steps progress from left to right.




                                          Figure 2        job Steps and Filter Modules



Digital Technical Journal     Vol. 3 No. 4    Fall 1991                                                                       47
Image Processing, Video Termj oals, and Printer Technologies




•    Printer status messages arc dispa tched to the           The VMS Printing SJ'Stem Environment
     operator and, in some cases, to the current user.        CPS funct ions as a component of the VMS printing
     CPS uses the normal V.\·IS OI'CO:vl notifica tion        system at the level of printer supervisor. As such, i t
     mechanism to send messages to the system opn­            interacts with, and is shaped by, the other compo­
     at or. If the user specified /NOTIFY in the print        nents of the VMS system . The term printer sup er­
     inst ructions, then CPS us<.:s the V.\IS SllRKTI IRt :   visor is used in this paper to be consistent with the
     syst<.:m service to send the message to the user         term inology of the em erging Interna tional Stan­
     also.                                                    dards Organ ization (ISO) Document Printing Appl i­
                                                              cation draft standard, ISO/IEC DIS 10175.
     In some cases, printer status messages require
     addi tional processing. for example. paper jams
                                                              Components
     requ ire spe<.:ial hand ling on some print<.:rs: since
     CPS cannot determine how many pages were lost            The V M S Batch/Print system is a general queue man­
     in th<.: jam, it invok<.:s human intervention by plac­   agement service, capable of queuing, schedu l ing,
     ing the job on hold. The operator or user can            and executing jobs in response to a variety of user­
     determine what parts of t he job, if any, to reprint.    specified instructions. 1 On the VMS system, the
                                                              printing instructi ons arc stored in a print job
•    Program status messages and user data messages           object, which is pl aced in a queue of jobs for a
     arc dispatched to the job log. If the user specified     printer. Modern print jobs often resemble batch
     /NOTIFY. then they are also displayed with the           jobs, due to complex stored processing instruc­
     $ ll R KTHRU system s<.: rvice. These messages may       tions and the heavy computing load placed on
     be printed or logg<.: d .                                graphics printer control lers.
                                                                 The V.\15 printing system contains components a t
   The input and output sides of the symbiont run
                                                              most levels o f the DECprint architectural model.
asynchronously most of the time, hut occasionally it
is necessary for the ou tpu t sid<.: to wa it for a mes­      •   l lser printing interface. The V M S system includes
sage from the printer. This synchronization between               interactive Digital Command Language (DCL)
the input side and output side of the symbiont is                 interfaces for printing and managing print jobs,
accompl ished by an internal event- signaling facil­              printers, and t he printing system i tsel f. 2 For
ity. When synchronization is r<.:qu ired, the output              DECwindows appl ications. the DECwindows Print
side wai t s for a specific named <.:vent and the input           Widget provides a graphical interface that per­
side signals that event when i t is detected . For                mits users to specify all the options for printi ng ,
example, a t the end of a job, CPS needs the final                a n d the ALL-IN-1 appl ication provides character­
printer sbeet_count i n order to calcu late the                   cel l menus for choosing print options, includ ing
sheet_ count for the job; this count is printed on the            the enhanced options offered by CPS.
trailer page and stored in the VMS accounting
                                                              •   job submission in terface. The VMS system
records. When CPS ne<.:ds the sheet_count, the out­
                                                                  includes program cal l interfaces that give the
put side waits for an event named sheet_count. The
                                                                  program a l l the capabilities of the DCL user
input side parses the incoming sheet_count mes­
                                                                  interface. 1
sage, stores the returned value in the shared data
structure, and signals the sheet_cou nt event. The            •   Print cl ient and service for remote printing. The
processing of this event is asynchronous: at the                  d istri buted queuing services product currently
time the message comes in, the output s ide may or                provides transparent remote printing in net­
may not have stal led while wa iting for the                      works using a proprietary network protocol.
sheet_count even£. If the output side was wai ting
                                                              •   Print spooler. The VMS job Control ler, recently
for that event, it is sched ul ed for fu rther process­
                                                                  replaced by the VMS Queue Manager, fu nctions as
ing; if the output side was not waiting, the event is
                                                                  queue manager and sched uler. (The function of
remembered, in case the output side attempts to
                                                                  spo oling printer data to temporary files is per­
wa it for this cond ition in the near hnw-e.
                                                                  formed by the VMS file system and is transparent
   In the next section we describe the ways CPS is
                                                                  to most components of the printing system.)
controlled and managed in the VMS printing system
and how it expands printing capabili ties in the VMS          •   Printer sup ervisors. The VMS system provides
environment.                                                      two standard symbionts to support most l ine




48                                                                    \kJI. j No. 4   Faii i'.J'Jt   Digital Techllica l jourtwl
                                            Design of the DECprint Common Printer Supervisorfor VMS Systems



    printers and serial printers. PRTSMB supports            job, but cannot specify certain other qualifiers such
    printers attached directly to communication              as DATA_TYPE. It can also permit users of old appli­
    ports on the CPU, e.g. , the printer port on a VAX       cations to access new fea tures of the printing
    workstation. LATSYM provides support for print­          system.
    ers attached to the serial or parallel ports of
    DECserver network communications servers. For            VMS Print Commands and Interfaces
    PostScript printers, CPS is used instead of these        The VMS printing system is manipulated through
    standard symbionts.                                      DCL commands and qualifiers. Many of the
  The VMS printing system also contains compo­               qualifiers are hand led by the queue m anager and
nents that affect CPS processing.                            have no impact on the operation of print sym­
                                                             bionts; others directly affect the operation of CPS.2
•   Device control libraries are col lections of sma ll      The VMS system also supplies a call interface to
    text sequences that can be inserted i nto the data       these functions.·'
    stream from the symbiont to the printer. The
    sequences are ideally organized into text l ibraries     VMS Interfaces to Symbionts
    containing named mod ules, with a separate
                                                             The VMS Job Controller/Queue Manager provides
    l ibrary for each type of output device . Device
                                                             two interfaces for customizing print symbionts: the
    control modules can be associated with a printer
                                                             PSM modu le-replacement interface, and the SMB
    queue by the system manager as part of a FORM
                                                             server symbiont interface. CPS is currently imple­
    definition or a job reset function, or accessed
                                                             mented as a single-stream symbiont through the
    d irectly by the user with the /SETUP qualifier.
                                                             SMB interface.
    Device control l ibraries frequently contain                The SMB interface perm its a user to replace the
    device-specific control sequences that alter the         flow of control of the symbiont with a separate pro­
    format of the text and pages, for example, setting       cess. The process may be writ ten in any style and
    printer paper margins, setting character pitch, or       structure suitable to the task a t hand, and need fol­
    enabling landscape printing. They may also con­          low only certain minor guidelines with respect to
    tain downloadable font data or preprinted data           the operating system environment. To use the SMB
    for each page.                                           interface, we replaced the entire symbiont process.
                                                             The result was much greater flexibility, but we
•   VMS form definitions contain page size and mar­
                                                             were required to write more program code.
    gin specifications that guide the print formatting
                                                                The SMB interface provides services to the sym­
    process for a print job. The user can also specify
                                                             biont process through subroutine entry points and
    page setup strings and can prohibit the symbiont
                                                             callbacks that pass messages between the symbiont
    from wrapping l ines during processing.
                                                             and the VMS queue manager. Messages from the
VMS Print Queues                                             system to the symbiont specify functions such as
                                                             start up, shut down, begin job, pause, resume, and
VMS has several distinctly d ifferent types of queues.
                                                             interrupt. Messages from the symbiont to the
Execution queues process jobs through a symbiont,
                                                             system return information such as job status, job
and generic queues transfer jobs to other queues.
                                                             completed, device status and error information,
Often generic queues are used for load balancing:
                                                             and checkpoint and accounting data.
one generic queue may feed several printers of sim­
ilar capabi l i ty and location.
   CPS also uses generic queues in an unusual way.
                                                             Range of Printers Supported
Default attribu tes can be specified for generic             CPS currently supports the fu l l range of PostScript
queues that cause all jobs subm i tted through the           printers supplied by Digital, from a low-speed
queues to inherit certain default print instructions.        color printer up to a 40-page-per-minute laser
For example, a queue can be established that, by             printer that can hand.le 11 different paper sizes.
defau lt, assumes that jobs are PostScript docu­
ments, or assumes that jobs should be printed in             Special I/0 Processing
landscape orientation. This abil ity to set defau lt         CPS supports several different means of com muni­
queue attributes is essential for supporting appl ica­       cation wi th the printer: serial, Ethernet, and a spe­
tions that can specify the queue name for a print            cial high-speed video connection.



Digital Technical journal   V!Jl. 3 No. 4   Fal/ 1991                                                           49
Image Processing, Video Terminals, and Printer Tec h nologies




     The serial connection may be either a d i rect con­     video connection or data bus. For such a controller­
nection between the compmer and the printer or               less printer to be general ly usefu l , the printing
a local area t ransport (LAT) connection by which            system must em u l a te a n existing class of prin ter.
printer is a ttached to a serial port of a DECscrver            The LN03 I m age printer (LN03Q) is a bit-map
terminal server. The two methods differ only i n             prin ter of this type. I t uses a special h igh-speed
t h e way johs are started and terminated . For              DMA bi t -map i nterface that p lugs into a Q-bus and
LAT-connccted printers, CPS must establ ish a nd d is­       provides the speed required for print i ng scanned
miss the LAT connection a t the start and end of             images. The protocol between this i n terface and
each job.                                                    the printer consists of bit maps and a sm a l l amount
     Once t he connection is establ ished with the           of status and synchronization information.
serial printer (via IAT or d i rect connect), CPS begins       The engine i ts e l f includes only the laser imaging
a dialogue with the printer using an asynchronous            and paper ha n d ling equipmen t. CPS hand les the
serial l ine protocol and PostScript programs. The           rest of the control ler fu nctions i n the host com­
asynchronous serial       I ine   protocol, defined by       puter. Because of the level of support and emu la­
Adobe Systems Inc . , consists of five control charac­       tion provided, the LN03Q printer appears to be an
ters that alter or query the state of the printer.           ord inary PostScript job printer with some special
     The symbiont forces the printer into an id le state     image capabi l i ties.
by a series of contro i/T, control/C, and controi/ D            For a given print job, CPS performs the normal
characters. When a controi/T resu lts i n an I DLE           processing up to the point a t which the PostScript
message from the printer, the symbiont and pri nter          languag�.: data stream would normally be sent to the
are ready to process a job.                                  printer. At this poi nt, CPS d i rects the data stream to
     PrintServer printers on Ethernet networks are           a special PostScript interpreter subrou tine that pro­
DECoct nodes. To write to a PrintServer prin ter, CPS        duces a bi t-map im age of the printed page in mem­
establ ishes a DECoct task-to-task session at the            ory The bit-map image is then sent to the prin ter
beginning of the job. The dialogue requi red for syn­        through a special LNV21 d i rect memory access 1/0
chronizing serial printers is not necessary for the          interface on the Q-bus.
Ethernet printers; the PrintServer protocols pro­              The software for the LN03Q printer a lso has one
vide synchronization ancl device control opera­              special processing path. The l.N03Q printer is
tions through separate control channels.                     in tended as an i m age printer for bi t-map images.
     Printers connected through Ethernet use several         CPS supports im age fi les containing page im ages
protocols, which are layered on DECnet task-to-task          that are scanned or precomputed at device resolu­
com munications. The protocol used depends upon              tion (300 clots per inch) and optionally compr<.:ssed
the version of the PrintServer code.                         with Comite Consu ltatif I nternationale de Tek­
     The local area print service (LAPS) protocol was        graphique et Tetephon ique (CCJTT) Group 3 (l D) or
developed fo r the PrintServer family and is sti l l in      Group 4 (20) compression methods. Image files can
use. The Common Prin ter Access Protocol (CPAP)              be transm it ted directly to the printer w ithout con­
wi ll replace LAPS in a l l PrintServer printers. ; PAP is   verting to PostScript. I m age files can only be sent
based on the earlier Reid-Kent protocol, I n ternet          directly to t he p rinter if they are printed one page
Socket 170, and is being discussed as a possible new         per sheet; if the user requests printing mu ltiple pages
Internet standard. '                                         per sheet, or other layup fu nctions, then the image
                                                             is processed through the PostScript in terpreter.
Special Processingfor "Dumb" Printers                          I mage fi les are structu red in Digital document
I n some printer configurations, i t is economical to        in terchange for m a t (DDI F), wh ich expresses text,
use the worksta tion or CPU as the printer con­              graphics, and images together. Files intended for the
troller. In t h is case, the printer i ncludes only the      LN03Q printer must contain only image bit maps.
print engine and media hand l ing and finishing                If the print job specifies DATA_TYVE= DDIF or the
equ ipment, ancl none of the electronics, comput­            file is a DDIF file, then CPS examines the file i n a spe­
ers, and interpreter programs that render the                cial mode. If the file correctly contains only image
graphics language into the elements requi red by the         b i t maps. CPS decompresses t he images in memory
print engine (usual ly a n array of pixels). Such a          if necessary, using the DECimage I mage Support
"dumb" printer is physical ly connected to the com­          Li bra11' rou tines. and then sends the u ncompressed
puter by a very high-speed l ink such as a direct            b i t map directly to the LN03Q print engine. Thus




50                                                                   !4JI. 3 Nu. 4    l-it// 1991   D igital Techn ical jou r, at
                                          Design of the DECprint Common Printer Supervisorfor VMS Systems



the image goes directly to the printer without pass­       Data Syntax Translation
ing through the PostScript interpreter.                    CPS provides a facil ity that translates selected
                                                           printer data syntaxes into the PostScript language.
Special Processing in CPS                                  The translating programs are subroutines, some
CPS includes a number of special features and func­        quite large and complex, that accept a data stream
tions to satisfy the requ irements of the DECprint         in one format and produce a data stream in another
architecture and the VMS printing system. In this          format. The translators are responsible for all for­
section, we d iscuss the features that extend the          matting, including sheet size, page orientation,
process of standard print symbionts or are com­            aspect ratio, and type sizes; CPS is responsible for
pletely new.                                               all I/0 and coordination with the printer. The trans­
                                                           lation facility currently supports the fol lowing
Reading Print Instructions                                 printer data syntaxes: DEC PPL3, ReGIS, Tektronix
CPS reads the print instructions for a job from the        4010/4014, and PCL Level 4.
VMS queue manager through the SMB$READ_                       The translation facil i ty has several restrictions. A
MESSAGE and SMB$READ_MESSAGE_ITEM functions                file may consist of only one data syntax, and all files
of the SMB interface. Print instructions are               in a job must be of the same data syntax.
expressed as attributes with values. Each attribute            In general, CPS performs the translation from
has an associated numeric code and symbol, called          one data syntax to another on the host computer.
an i tem code, and a value of a specific data type.        In this way, simple printers that support only the
The symbiont reads each item code and value, and           PostScript language interna l ly can be extended
stores the information in a static data structure.         to support a number of printer languages. This
The information is used later to determine the pro­        reduces the requ irement for a complex printer con­
cessing sequence for the job, special information to       troller that supports m u ltiple data syntaxes inter­
be displayed on separator pages, and so forth.             nal ly. Host translation can guarantee consistent use
                                                           across jobs of the printer's internal fonts, page ori­
Bidirectional Communication with                           entation, finishing equipment, and page layup The
PostScript Printers                                        general mapping of page images to sheets supplied
CPS requires a fu l l duplex communications path to        as part of CPS requires that the printer operate in
PostScript printers since they report many condi­          PostScript mode. To ensure consistent use of fonts
tions by sending messages to the host computer.            and consistent positioning of pages with respect to
These messages include device status messages,             finishing such as duplexing and stapling, all lan­
program status and error messages, user data mes­          guage translation must be done by the symbiont.
sages, and replies to CPS inquiries.
   CPS also requests information from the printer          Page Layup Multiple Pagesper Sheet
for synchronization, formatting, and accounting            Page layup is the process of printing more than one
purposes. For instance, to determine how to for­           page image on a sheet of paper. When more than
mat ANSI text, the symbiont needs to know what             one page i m age is placed on a sheet of paper, the
paper is loaded in the printer.                            i m ages are rotated and scaled to fit on the page, but
   CPS receives the messages from the printer and          are altered in no other way. The layup facility works
parses them to determine what it should do with            with all data types, includ ing PostScript and PCL
the message. If the message is device status, then         data syntaxes. Layup also permits formatting for
CPS routes the message to the operator and/or the          larger paper sizes and then printing on smal ler
user whose job is being printed. If the message is an      sheets.
internal CPS communication, then CPS processes it.             Layup is invoked expl icitly with one or both
Otherwise, the message is e ither a program status         of the extended qualifiers NUMBER_UP and LAYUP_
message or a user data message. In either case it is       DEFINITION. NUMBER_UP specifies the maximum
Jogged for the user.                                       number of page images that will be printed on a
   Al l messages are parsed except user data mes­          single side of a sheet; for example, two-up printing
sages. Messages from the printer's interpreter are         is specified by the " NUMBER_UP = 2 " option. Two or
converted to a standard format that would, if              four page images per side may save significant quan­
desired, permit the message to be translated i n to        ti ties of paper for draft printing, handouts, and the
the user's native language.                                l ik e . Up to 100 page i mages may be placed on a



Digital Technical jounwl   Vol. 3 No. 4   Fall 1991                                                              51
I mage Processing, Video Terminals, and Printer Tec hnologies




single sheet of paper for thumbnail d raft printing to          PostScript environment, including the alt<.:rctl p:1ge
review the overal l l ayou t of a document.                     geometry, e.g., layup established by the print j o b .
   Layup may also he invoked thro u gh a combina·                  CPS defines two n e w separa tor pages. T h e fi le
tion of PAGE_SIZE and SHEET _SIZE with NUMHER_UP.               error page is print ed when a tile cannot be opcn�.:d
for example, the combination of PAGE_SIZE = E,                  o r an error occurs while r�.:ad ing the file. The file
SHEET_SIZE = A , NUMBER_UP= 1 perm i t s printing               error page informs the usn of the error cond ition
draft cop ie s of large-format documents on small               which caused i t to be print<.:d . The job log page con·
paper. Conversely, the combination of PAGE_                     tains up to 40 l ines of the job log fi le. The job log file
SIZE =A,SHEET_SIZE = B,NU.\1 BER_U P= 1 magnifies the           contains j o b events such as job start and job com·
smal ler page to tit the larger sheet.                          pletion as wel l as program status messages and user
                                                                data returned from the printer.
Duplex Printing
Pri n t i ng on both sides of the paper i ntroduces a           Managing Printer Resources
numbcr of n e w options and interact ions that                  Once com mun ication is established with the
req u i re spe ci al processing in C P S . CPS begins each      serial printer, the symbiont must esta blish what
document on the first siclc of a new sheet, so that             resources are available on the printer. These
recto and vcrso (right-hand and left-hand) pages                resources include prologues, which are comm only
and a lte rna t ing margins arc a l igned with the cor·         used Post Scrip t rou t ines, the amount of available
rect sides of sheets as they arc stacked by the                 virtual memory, and the med ium in the default
prin ter. This function :1lso interacts with the d i r<.:c·     paper tray. For example, CPS persistently loads the
tion in which the medium is physi cal ly loaded i nto           PostScript prologue fo r the output of th e ANSI text
the printer if the medium is not symmetric left-to·             translator into the PostScript interpreter. Th is
right, top- t o - bot t o m , or front· tO·back, such as pre·   resou rce might be lost to the printer because of a
drilled paper.                                                  power fai l u re or might become obsolete clue to a
   The interactions of POL coordinate systems, page             software upgrade. CPS interrogates the pr i n t er at
layup, media selection, asymmetric media, duplex                the beginning of any job req u i ring the translator
p ri nt i ng and binding are the most elusive engineer·
          ,                                                     prologue and loads a new prologue, if necessary.
ing problems in t h e printing :1 p pl ic a t i on space. No    CPS also performs similar processing for the
gennal model of these interactions h as been devel·             PostScript prologue that is used to generate the
oped, despi te considerable effort in standards com·            separator pages.
mittees. I t appears that i t is necessaq• to implement           For trad itional resources such as paper, CPS relies
every p oss i ble option.                                       on status messages from the printer to indica te that
                                                                the pr i nte r is stopped because paper supply is
Separator Pages                                                 empty or j ammed. These condi tions are relayed to
CPS prints a l l the separator pages defined by th e
                                                                the operator and to the current user by s ta ndard
                                                                VMS mechanisms.
VMS queuing system as wel l as some generated by
CPS. Fl a g, burst, and trailer pages for j o b and file lev­
els are ava i lable as defined by VMS, and contain              Library Search Lists
the same informat ion presented in a hi gh l y lcgible          I n the standard VMS print symbiont, only one
format. In addition to the standard V:'v!S infor·               device control l i brary may be associated with a
mation, the j o b trai ler page also conta ins the first        queue. This is not a probl.em since the standard VMS
two PostScript language errors returned from the                print symbiont deals with only one data syn tax .
printer. This often makes it unnecessary tO use                 (Recall that clevice control l i braries are often writ·
M ESSAGES= PRJNT to see simple errors.                          ten in device-dependent data syn tax.) CPS, on the
   To ensure that the job separator pages can always            other hand , uses more than one data syntax when
be printed correctly, CPS r<.:s<.:ts the POL interpreter        printing a non-PostScript job: the data stream to the
in the printer befor<.: printing these pages. The CPS·          printer is PostScript, but the data stream ro tbe
generated scparator pages do not alter the coordi·              translator is in another data syntax.
nate system of the interpreter; the user's document                Early versions of symbionts that supported
starts printing with the default Pos t Script state. File       PostScript su ffered from the same restriction: only
separator pag�.:s, in contrast, print in the current            one device control l i brary was ava i lable, and i ts




52                                                                     Vol. 3 No. 4   Fall 1')')1   Digital Tecbnicaf jou ,..,w l
                                               Design of the DECprint Common Printer Supervisorfor VMS Systems



modules were expre�seu i n PostScript. This made i t             on this article. F inal ly, we thank the many si tes and
impossible for users t o share device control                    p eople who have tested the DECprint Printing
l ibraries with their stand ard VMS print symbiont               Services software.
ami their non-PostScript printers.
     To solve the problem of multiple data �yntaxes              References
in a job. CPS introduced device control l ibrary
search lists. The system manager, rather than speci­             1 . VMS Utility Rou tines Manual (Maynard: D igital
fying a single file specifica t ion in the I N!TlALIZE/             Equipment Corpora tion, Order No. AA-LA67B-TE,
QUEUE/LIBRARY com mand. creates a logical name                      1990).
instead . CPS transla tes that specific logical name             2. VMS DCL Dictionary, 2 vols. (Maynard: D igital
and uses each element of the resu lt as a device con­               Equipment Corporation, Order Nos. AA-PBK 5A-TE
trol l i brary. Each l ibrary in the search li st can have a        and AA-PBK6A-TE, 1991 ).
data syntax associated with i t by adding the
qual ifier, /DATA_TYPE = .                                       3. Vil,.JS System Services Reference (Maynard: D igital
     CPS    suppl ies a device control               l ibrary,      Equipment Corporation, Order No. AA-LA69A-TE,
CPS$ DEVCTL, wh ich must be included in the search                  1991 ).
 l ist, usual l y as the first, or on ly, element in the         4. ]. Jones, A. Kachran i, and T. Powers, "The
search I ist.                                                       Common Printer Access Protocol," Digital
                                                                    Technical journal, val. 3, no. 4 (Fall 1991 , this
Summary                                                             issue): 55-60.
The DECprint model of prin ting describes a useful
                                                                 5. B. Reid and C. Kent, "TCP/IP PrintServer Print
structure w i th consistent functions and responsi­
                                                                    Server Protocol," Western Research Lab Technical
bil ities. CPS is an advanced print symbiont that runs
                                                                    Note TN-4 (Maynard: D igital Equ ipment
in the VMS printing system. I t includes many spe­
                                                                    Corporation, 1988).
cial ized functions to support the features of a wide
range of modern printing devices. It provides, we
feel, an extraordinary level of support. It was                  General References
designed with a highly modular and flex ible inter­
nal structure to permi t enhancements to be engi­                Guide to Maintaining a VJ11S System (Maynard :
neered with minimal interactions with current                    D igital Equipment Corpora t ion, Order No.
operations.                                                      AA-LA34A-TE, 1990).
   CPS is currently shipping its fourth version. Th is           DECprint Printing Services User's Gu ide (Maynard:
version fu l ly supports the ten d ifferent PostScript           D igital Equ ipment Corpora tion, Order No.
printers suppl ied by Digi tal. which range from a               A A-PBZGA-TE, 1991) .
low-speed color printer to a h igh-speed laser
prin ter. It also supports five different data syntaxes          DECprint Prin ting Services System !Hanager's
in which appl ications can wri te documents. We                  Guide (Maynard : Digital Equ ipment Corporation,
expect that more printers and more capabil i ties                Order No. AA-PBZFA-TE, 1990).
will be added in future versions, and that CPS wil l             Digital ANSI-Compliant Printing Protocol Level 3
require a minimum o f add i tional engineering effort            Programming Reference Manual (Maynard: D igital
due to its very general internal structure.                      Equipment Corporation, Order No. EK-PPLV3-PM,
                                                                 1991) .
Acknowledgments
                                                                 Digital ANSI-Compliant Printing Protocol Level 3
We would l ike to thank Peter Conkl i n for actively
                                                                 Programming Supplement (Maynard: Digi tal
initiati ng CPS and Gary L. Brown for even more
                                                                 Equipment Corpora tion, Order No. EK-PPLV3-PS,
actively expanding i t . We would also like to thank
                                                                 1991) .
past and cu rrent CPS developers: Ned Batchelder,
Cathy Cal lahan, Mark DeVries, Rich Emmel, Dave                  PostScript Translator's Reference Manualfor ReGIS
Gabbe, David Larrick, Klara Levin. Mary M arotta,                and Tektronix 4010/4014 (Maynard: Digital
Doug Stcf:tnd l i , and Charlotte Tim lege. We would             Equ ipment Corpora tion, Order No. AA-PBWFA-TE,
like to thank Bill Fisher fo r h is extensive com ments          1991) .




Digital Techuicaf ]ouruaf      Vol. _; No. 4   Pall 1991                                                              53
Image Processing, Video Terminals, and Printer Technologies




PostScript Printers Programmer's Supplement
(Maynard : Digital Equipmen t Corpora tion, Order
No. EK-POSTP-PS, 1991 ).

PostScript Language Reference Manual, 2nd ed . ,
Adobe Systems Incorpora ted, ISBN 0-201-18127-4
(Reading, MA: Addison-Wesley, 1990).

Information Technology- Text and Office Sys­
tems-Document Printing Application (DPA),
ISO/IEC JTC 1/SC 18 N, Draft International Standards
10175-1 and 10175-2 (September 1991 ) .

CDA Base Services Technical Overview (Maynard :
Digital  Equipment      Corpora tion,     Order   No.
AA-PHJYA-TE, 1991 ).

Creating Compound Documents Using CDA Base
Services (Maynard: Digital Equipment Corporation,
Order No. A A-PHK2A-TE, Il)89).

Writing Converters Using CDA         Base Services
(Maynard: D igital Equipment Corporation, Order
No. AA-PHK 1 A-TE, 1991 ).

CDA Base Services Reference Manual, 2 vols.
(Maynard: Digital Equipment Corporation, Order
Nos. A A-PHJZA-TE and AA-PHKOA-TE, 1991).

CDA: DDIF Technical Specification (Maynard : D igital
Equipment Corporation, Order No. A A-PHK3A-TE,
1991) .

('DA: DTIF Technical Specification (Maynard: Digi tal
Equipment Corporation, Order :-.Jo. A A-PH K4A-TE,
1991 ).

DDIS   .�vn tax Spec�fication (Maynard: Digital
Equipment Corporation, Order No. EL-00081-00-1 ,
1987).




54                                                            Vol. .3 No. 4   Fa/1 1')91   Digital Tecbnical ]ournal
                                                                                                  james D. jones
                                                                                                 Ajay P. Kachrani
                                                                                               Thomas E. Powers




The Common Printer Access Protocol
               The DEC PrintServer Supporting Host Software version 4.0 incotporates Digital's
               first implementation of the new common printer access protocol (CPAP). This pro­
               tocol is compatible with the local area print server (LAPS) protocol, which was
               optimized for VMS access and DECnet transport, and with tbe Reid-Kent proto­
               col, a PostScript-based, TCP/!P-connected print server for a client-server environ­
               ment. The CPAP protocol supports a variety of data presentation protocols
               and allowsprinters to be connected to driving applications by various communica­
               tions and process-to-process intetfaces. Tbe protocol also couples entities running
               different operating systems across disparate networks. Because of its superior
               performance, the new CPAP protocol has been accepted by the Open Software
               Fou ndation for inclusion in a future release of OSF/1.



The presentation of computerized data has become             cal prin ting device. The comm o n printer access
a remarkably sophisticated and subtle operation.             p rotocol (CPAP) described in this paper provides
Video displays now support windows with com­                 the fundamental services required by a printer
plex al locations of display space, variable fonts, and      supervisor for the presentation of data and col lec­
real-time user input operations. Printing devices            tion of accounting information. In addition, the
now offer support for publication-quality fonts,             CPAP supplies easier network access between
line art, and images. These devices can present              printer supervisors and printers, as well as ancil­
visual objects on a variety of media, from many              lary control of p rinters for network management
sources, and in variable orientations and presenta­          and device configuration. The CPAP also provides
tion modes. In addition, both video and printing             services to d istribute the processing requ irements
devices are now decoupled from dedicated com·                of the printer i tself, most notably a mechanism for
puting environments, and are shareable from many             del ivery of network fo nt services. This last capabil­
hosts and by many users or programs.                         ity al lows a printer to offer what amounts to vir­
    Now, only the simplest printing devices are l im­        tual services, i .e . , the abi l i ty to configure i tself
i ted to presenting just characters, and many users          dynamically to the demands of a print job without
are finding such restricted capabi l ities inadequate.       the involvement of the printer supervisor.
Also, most printing devices still requ ire dedicated            This paper begins with a discussion of the
connections to single computers. However, more               influence of existing protocols and the DECprint
printers now offer fu ll network accessibil ity; i.e. ,      architecture on our CPAP design goals. The sections
network printers are capable of offering sophisti­           that fol low present the printer session concepts
cated services to a wide variety of users and their          and the functional i nterface between the protocol
applications.                                                and appl i cations. We then describe the implemen­
    The paper entitled " Design of the DECprint              tation of the new protocol in a server environment,
Common Printer Sup ervisor for VMS Systems"                  including interoperabili ty, compatibility, and the
in this issue of the Digital Technical journal               translation of the older PrintServer protocol. At the
describes access methods and interrelations                  close of the paper, we discuss ongoing standard­
among services that provide for these increasingly           ization issues.
sophisticated data presentation capabil i t ies. 1 The
printe r access protocol (PAP), a service interface in       History
the DECprint arch itecture, couples the printer              The PrintServer 40, D igital's first ful ly networked
supervisor component to the logical printer for              pri nter, was first shipped in 1986. Its local area print
presenting data and otherwise controlling a physi-           server (LAPS) protocol was analogous to later



Digital Technical journal   Vol. 3 No. 4   Fa/1 1991                                                                55
Image Processing, Video Terminals, and Printer Technologies




printer access protocols. The PrintServer 40 was a         problem was a major conceptual test in the first
ground-breaking p roduct for D igital, and the LAPS        implementation of a CPAP server. Th is is discussed
protocol was a major aspect of the PrintServer             in more detail in the sect ion The CPAP Server
development effort, portions of which date back to         Implementation.
198:1 . The LAPS protocol was designed and devel­              The Reid-Kent protocol met many of the techn i­
oped with particular product-oriented deliverables         cal design requirements for a new PAP . It was built
in mind, and was optimized for VMS access and              on industry-standard components, and contained
DECnet transport. While this protocol predates             no proprietary technology that wou ld prevent its
much of the architectural work now being imple­            publication.
mented in Digital's printing products, it was (and             However, certain PAP design goals were not cov­
still is) a signi.ficant element of PrintServer archi­     ered by the Reid-Kent protocol in its 1988 version.
tecture and implementation.                                •   There was no faci l i t y to select a specific page
     Work began on more general PAPs in 1987 as
                                                               description language (PDL) for printers support­
part of the early work on the D ECprint archi tec­
                                                               ing multiple interpreters.
ture (known at the time as the Printing Systems
Model). The specifics of what would become the             •   There was no method for soliciting the capabi l i­
CPAP emerged in late 1988 in two internal papers               ties and media available on the printer.
by Brian Reid and Chris Kent of Digi tal's Western         •   The    only    language        supported     was    English
Research Laboratory. These papers presented the                (contrary      to    the    corporate      guidelines   for
initial design concepts for a PostScript-based ,               international ization).
TCP/IP-connected (transmission control protocol/
                                                           •   Data sent from the printer was not ca tegorized ;
internet protocol) print server in a clearly defined
                                                               user-specific information was mixed with opera­
c l ient-server environment. This print server proto­
                                                               tor and service data.
col came to be known as the Reid-Kent protoco l.
                                                           •   No means was provided to solicit the status of
                                                               the printer.
Design Rationale and Goals
By early 1988, design goals for (and constraints on) a     •   There was no encoding to discriminate between
PAP were wel l u nderstood, and had been col lected            binary and text files.
and publ ished as part of Digi t a l 's Printing Systems       However, these flaws were largely omiSSIOns
Model. Chief among these goals and constraints was         from the design goals, not fundamental conflicts
the need to support a variety of data presentation         with them. The archi tecture team decided that the
protocols, and to al low printers to be connected to       Reid-Kent protocol could be extended to add ress
driving appl ications by a variet y of com mu nica­        these omissions without serious conflict. I n fact,
tions and process-to-process interfaces.                   the necessary extensions were designed to al low
  The i ncreasing corporate comm i t ment to open          cl ients and servers conforming to the original Reid­
systems made i t clear that a PAP would also have to       Kent protocol to remain in conformity with the fu l l
couple entities runn ing various operating systems         C:PAP specification.
across different networks. Thus, the DF.Cprint PAP
architecture team decided early in the design pro­         Architecture
cess that a PAP should be designed for public              The CPAP is primarily a com m u n ication-orien ted
access: that is, t he specification for the protocol       protocol, i.e. , the presentation of its function is
should be put into the public domain and submit­           closely coupled with its encoding. The major syn­
ted for industry standardization.                          tactic fea tures of the < : PAP derived from the Reid­
     Jnteroperability is a most serious constraint.        Kent protocol are the fol lowing.
Digi tal has a strong tradition of maintaining back­
                                                           •   All encoclings are ASCII strings. This eases the
ward compatibility within and among its product
                                                               generation of protocol streams and ensures inde­
fam i l ies. I n a distributed processing environment,
                                                               pendence from the u nderlying comm u n ications
however, backward compatibil i t y takes on the
                                                               channels.
added burden of i nteroperabil i t y. Mult iple clients
must com m unicate with mult iple servers. any of          •   No data fields are fixed length. This provides for
which can be u pgraded to new versions of sup­                 extensibi l i ty of the protocol and eases the gener­
ported pwtocols asynchronously. Addressing this                ation of a protocol stream .



56                                                                   Vol. 3 No. 4   Fall I'J'J/   Digital Technical journal
                                                                        The Common Printer Access Protocol



•   Multiple channels of communication use the            will use for each document and which virtual cir­
    same basic format. Common parsing of separate         cuit will be used for its transmission.
    channels simplifies implementations.                     Selection of the proper virtual circuit for trans­
                                                          m ission of documents to the printer is performed
•   Simple numeric tokens define the operators.
                                                          by passing tokens from the printer to the printing
                                                          service. The tokens are then mapped to whichever
Session Concepts                                          v irtual-circuit service is being used by both the
The CPAP architectu re defines separate contexts for      printing service and the print server. This map­
each type of work the CPAP can perform. Each con­         ping approach avoids passing network-specific
text requires that a separate session be established      information within the protocol. Not only does the
for its own tasks, and each session involves the cre­     approach make the CPAP independent of the net­
a tion and use of a separate network connection           works on which i t might run, it ensures that the
between the controll ing cl ient and the server. Each     network services need no knowledge of CPAP
connection identifies the type of session the initia­     encodings. Such v irtual-circuit m apping is criti­
tor requires. The CPAP defines three d ifferent ses­      cal to allow CPAP cl ient-server processing to be
sion types: print, management, and console.               implemented in a heterogeneous, internetworking
    The set of CPAP operators al lowed for a session is   environment.
restricted to those needed to support that type of            During the printing of the document, some data
session. All session types have access to printer         presentation interpreters (PostScript, for example)
status and configuration information. In addi tion,       send data back to the user or print service. In addi­
mu ltiple concurrent sessions are permitted. Print        tion, the printer may run out of paper or toner,
sessions and management sessions may have one or          may have a ful l output tray, or may encounter other
more virtual circu its active to a printer at a time.     exception conditions not directly related to the
The use of mu ltiple circu its permits the streaming      interpretation of page description data. The CPAP
of data to the printer over logically separate chan­      categorizes such conditions and delivers relevant
nels, thereby eliminating application protocol over­      messages to the user, the operator, or the event logs.
head for the most frequent operations. In contrast,           Upon completion of the job, the printing service
console sessions use a single virtual circui t for        is notified of the media used , the number of pages
exchange of data with remote terminals.                    printed, and the printer processing time requi red
                                                          to complete the job. The protocol also includes a
Print Sessions Print sessions usually consist of a        provision to abort jobs, e.g., an improperly formed
series of documents printed for a user on a given         document that might otherwise hang the printer.
host by a printing service (a "printer supervisor"
as defined by the DECprint architecture). With the        Management Sessions The CPAP supports certain
operators provided by the CPAP, the printing ser­         printer services through management hosts. A man­
vice can determine the language interpreters,             agement host is a network entity (not necessarily
printer options, fonts, prologues, and media that         the same entity as the printing service) with which
are currently instal led at the server. These opera­      the printer can exchange i nformation or request
tors also provide the current operational state,          services. Such services include
number of jobs queued to the printer, and the cur­
                                                          •   Time service
rent job status. These features permit the printing
service to select the printer (server) that can satisfy   •   Centralized event logging
the user's request and to determine a method for
                                                          •   Centralized accounting
submitting the job to the printer.
   Once the printing service has begun a session          •   Program loading and configuration
and identified itself, it identifies the user and the
                                                          •   Font services
user's job code to the printer. This information may
be used by the printer to provide usage information         An important aspect of the CPAP is that the
to a centralized accounting service. The printing         printer is always passive with regard to initiating
service can then present documents to the printer.        management services. A cand idate management
A transaction between the printing service and the        host advertises that it has services to offer, and a
printer establ ishes which interpreter the printer        print server accepts or rejects the offer. Once a




Digital Tecbnicaljourual   Vol. 3 No. 4   Fall 1')91                                                         57
Image Processing, Video Terminals, and Printer Tech nologies




connection with one or more management hosts               programm ing interface (API) that a i i O\:vs access to
is established, the printer may use such hosts as          all CPAP facil ities is included in the protocol's
servers for time synchronization, configuration file       specification.
access, and font lookup. Additional functions for              A connection block, which is passed as a parame­
these hosts may be loading program images, event           ter to all functions, provides support for vari­
logging, accounting, and general file access.              ous printer types, their device identifications, and
    File naming to access general file services is a       descriptors for command and data channels. This
problem that needs special attention if the server         support includes separate com mand and data
and the protocol are to maintain independence              channels for printers supporting mu ltiple virtual
from the host operating systems. Commonly used             circuits or channels. Just as in the case of the data­
files a n: identified in the CPAP by reserved tokens,      stream form of the protocol, the API form allows
such as $CONFIG, $ DEFAULTS, $ RESOURCES, and              separate channels for data and comm ands.
$SETUP. Arbitrary path names are allowed, but can              A separate command channel allows ease of con­
access only a limi ted domain (from a known root           trol flow between client and server. This may
directory) to preserve file system independence            include the client receiving the server's status or
and to maintain security.                                  events, or the cl ient sending aborts to the server.
   Translation to the host's services is provided          For devices that support only a single channel, the
by the host i tself. This permits the printer to be        generic printer driver can set both command and
served by d ifferent hosts using a wide variety of         data channels to the same valu e . For supporting
operating systems (and their implicitly different          mu ltiple jobs active at the same time (job overlap),
file-naming conventions and syntaxes) without any          a job identification (ID) parameter is passed with
awareness of a management host's implementation            all functions.
by the server.                                                To support various message types, the address
                                                           of a read-ca l l back routine is passed to the open
Console Sessions A console session is a form of            printer function along with a pointer to read-cal l­
printer management. The content of the data                back arguments. These arguments may signal vari­
exchanged du ring a console session is specific to         ous events, or may consist of messages for the user,
the printer, and is not specified by the CPAP.             operator, accounting, or resources available in the
Services performed within a console session might          printer.
include                                                       An early version of the generic fu nctional inter­
                                                           face was part of MIT Project Athena's Pallad i u m
•    Operator services, such as tel ling a printer what
                                                           Print System. The printer supervisor in D igita l 's
     media have been loaded (e.g., by color, weight,
                                                           LN03R ScriptPrinter product was modified to cre­
     or transparency), or sett ing physical printer
                                                           a te a generic printer i n terface for both the
     defaults (e.g. , duplex versus si mplex, or default
                                                           ScriptPrinter device and the PrintServn fam i ly.
     medium selection)
                                                           This conversion from an API-accessible base took
•    Network management configuration services,            one week to execute, whereas it typically takes
     such as control I ing domain access to or from        six months of effort to develop a new printer
     the printer                                           supervisor for a device as complex as the
                                                           PrintServer product.
•    Troubleshooting or debugging services

  D igital's implementation of console services on         The CPAP Se rver Implementation
cu rrent PrintServer products conforms to the              The implementation of a protocol gives rise to
Enterprise Management Architecture.                        problems different from those related to its design.
                                                           When defining the archi tecture, one strives to pro­
Application Program Interface                              vide an ideal that includes all of the desired features
The fu nctional interface to any protocoJ provides         in an elegant manner. When performing an imple­
an additional abstraction between an application           mentation, one finds that elegance often has to take
and a protocol. This abstraction answers many of           a back seat to pragmatics. This is especially true

today's software appl ication needs, including inter­      when the new protocol is intended to replace two
operability, portability. modu larity, and reusabil­       different protocols in a new version of an existing
ity across mu ltiple architectures. An applica tion        product. Merely implementing the new protocol



58                                                                V!J!. 3 No. 4   Fall 1991   Digital Tee/mica/ journal
                                                                      The Common Printer Access Protocol



is not enough-the implementation must some­             biont (CPS), several versions of the U LTRIX oper­
how coexist with the protocols being replaced.          ating system, and a source kit version running on
   D igital's first production implementation of the    a Sun Microsystems workstation.
CPAP was targeted for the DEC PrintServer Sup­             Unfortunately, this testing uncovered latent
porting Host software version 4.0, which loads and      defects in the implementation of the existing prod­
drives the PrintServer fam i ly of printers. For the    ucts. We had to analyze each of these defects and
rest of this paper, we refer to this software by the    plan corrective action. Since updating the existing
PrintServer product designation of LPS version 4.0.     products in the field is a difficult process (both
   We started the implementation by modifying           technically and procedurally), we corrected most
D igital's ULTRJX PrintServer client, which already     of the defects by altering the server to deal with the
used the Reid-Kent subset of the CPAP, to use           problems. Retesting was performed over several
DECnet network transport and run on the VMS oper­       baselevels to ensure that our changes caused no
ating system. We then updated the LPS server            regression.
code to permit ei ther DECnet or TCP/IP transport.         At one of the early baselevels, the interface
This was accomplished by using the direct -to-port      between the network distribution software and the
communication features of the VAXELN operating          server's PostScript interpreter was updated to use a
system. The server establ ishes a circui t u sing the   stream-based connection in place of the previous
appropriate transport and then spawns a process         packet protocol. This update permitted the new
for deal ing with each incoming connection. Thus,       CPAP data channel to be mapped by reference to
the same code can service print sessions, manage­       the input of the PostScript POL or any other POL
ment sessions, and console sessions without con­        supported by the printer. This change alone per­
cern for the type of network transport.                 mitted the performance of the server to be main­
    The CPAP was, by design, directly upward­           tained even when the server was translating from
compatible with the Reid-Kent protocol subset.          the old LAPS protocol to the CPAP.
However, D igital's PrintServer offerings prior to         In general, development proceeded incremen­
LPS version 4.0 were LAPS-based, and LAPS was not       tal ly, i . e . , key features were identified and added
CPAP-compatible. To permi t users of existing           with each baselevel. While this technique limits the
PrintServer printers to continue to use these           complexity of prod ucing the product, it raises an
 products, we had to find a way for the new CPAP        important business issue. Specifically, the provi­
 implementation to coexist with the older LAPS          sion of enhanced services in a cl ient-server envi­
application protocol. We achieved this coexistence      ronment often exposes aspects of the proverbial
by having the server p erform translations from the     " chicken-and-egg" situation. There is l ittle call to
older protocol to the new one in the server i tself.    offer enhanced features in a server if clients have
When the client establishes the initial connection,     not been programmed to solicit the features. How­
the server senses which protocol is being used by       ever, cl ients are not readily upgraded to solicit
the client system. If the initial message indicates     features that might not be widely available.
the use of LAPS, the server spawns i ncoming and           The LPS version 4.0 project team met i ts backward
outgoing filters to deal with the incoming connec­      compatibility design goals by including the LAPS-to­
 tion, and a new internal circu i t replaces the        CPAP filters. In doing so, they undercut the need
network connection to handle the interpretation         to provide the enhanced feature support that the
of the CPAP.                                            CPAP was designed to deliver, since existing cl ients
    The coding of the LAPS filters was the last step    (earlier versions) coul d not avail themselves of
in implementation before testing began. The             the added features. In addition, the risks of includ­
PrintServer 20, PrintServer 40, PrintServer 40 plus,    ing fu l l CPAP support in LPS version 4.0 (possible
and the new turbo PrintServer 20 all had to be          increase in time to market, and the creation or expo­
 tested using both LAPS and the Reid-Kent subset of     sure of more latent defects in all supported environ­
 the CPAP. In addition, the new implementations of      ments) seemed to outweigh the benefits. However,
the management cl ient and the console cl ient on       a last-minute change to use the new protocol's data
the VMS system requi red verification. This verifi­     channel for loading fonts yielded such a large per­
cation entailed a mu ltitude of tests using the LPS     formance advantage that resistance to using the
symbiont running on older versions of the VMS           new features crumbled , and the project team was
operating system, the newer common print sym-           allowed to submit the fu ll protocol to field test.



Digital Technical journal   Vol. 3 No. 4   Fall 1991                                                         59
Image Processing, Video Te rminals, and Printer Tech nologies




Standardization                                           architecture and created the first prototype imple­
Network printing became widely available i n the          mentations. Jim jones championed the proto­
 mid-1980s, but products from different vendors           col in the DECprint PAP archjtecture team (Alan
were not compatible. Network printing protocols           G uenther, Tom Hastings, Jim Jones, Tom Powers,
were largely proprietary efforts by vendors who           and Eric Rosen) and coded the LPS version 4.0
had developed them for their own printer prod­            server. Carol Gal lagher wrote the LAPS fi lters to
ucts. Digi tal's PrintServer 40 and i ts LAPS protocol    translate from the old protocol to the new. M ike
were typical in this rega rd . By the late 1980s,         Augeri and John McLain ported the management
 network printing was an establ ished and competi­        and console clients to the VMS system from the
tive technology, but there was still l i ttle i nter­     U LTRI X system . ]. K. Martin rewrote the Berkeley
operabil ity among the various vendors' p roducb.         Software Development (BSD) sou rce kit to use the
    In the absence of printing protocol standards, the    new protocol. Ajay Kachrani developed our ULTRIX
Internet Engineering Task Force (I ETF) formed a          and M IT Athena clients and represented the proto­
Network Printing Protocol working group in                col d uring early phases of the IETF standardization
early 1990. This group's charter was to examine           effort. Many others supported these efforts, and
prin ting protocols then in existence or u nder devel­    others are yet beginning to develop new CPAP
opment, assess their appl icabi l i ty to Internet-wide   clients. We thank them all for their efforts.
use, and suggest cha nges. Digital's representatives
to the I ETF working group on the Palladium               Reference
Printing Systems standardiza t ion reported the inter­
                                                          1 . R. Landau and A. Gu enther, " Design of the
est shown in D igital's Reid-Kent protocol. Thus, in
                                                              DECprint Common Printer Supervisor for V M S
July of 1990, Digital submitted a version of the PAP
                                                              Systems," Digital Technicaljournal, vol. 3, no. 4
that was u nder consideration by the DECprint PAP
                                                              (Fall 1991 , this issue): 43- 54.
archi tecture team.
   Early consideration of this PAP by IETF and the
LPS version 4.0 implemen tation effort ran concur­
rently. This provided a unique opportunity for
Digi tal's implementers to obtain feedback from a
very knowledgeable architectural communi ty. In
turn, they could report implementation experi­
ences that affected the review and progress of the
specifi cation towards standardization. Implemen­
ta tions of CPAP clients and servers by companies
other than Digital are in progress.
   As part of Project Athena's Palladium Printing
System, the CPAP has been accepted by the Open
Software Foundation for inclusion in a future
release of OSF/ 1 .
   A draft of the CPAl' i s being circulated among
Internet members for com ment. Meanwh ile, work
on future enhancements cont inues. Wo rk is now in
progress to specify a superset of the existing pro­
tocol that deals with a u thentication and encryp­
tion to strengthen security. This work is being
done in the spirit of the original migration from the
Reid-Kent protocol to the CPAP; i.e., the security
features being added will not adversely impact
users who do not need the new features.

Acknowledgments
The CPAP effort has been the work of many devel­
opers. Chris Kent and Brian Reid drafted the base



60                                                              VIii. 3 No. 4   Fall 1991   Digitttl Technical journal
                                                                                                  Guido Sim one
                                                                                            jeffre y A. Metzg er
                                                                                                  Gary Vaillette




Design of the Turbo
PrintServer 20 Controller

               The turbo PrintServer 20 controller is a pe1formance enhancement of the original
               PrintServer 20 system controlle�: The turbo controller was developed to enable
               PostScript code to execute faster and thus improve page throughput for complex
               documents. The RETrACE analysis system was designed to analyze tbe pei formance
               of the original PrintServer 20 system and estimate expected performance future
               systems. Tbe turbo controller's processor and its three subsystems for memory,
               write buffer; and bit-map data transfer were selected based on the analysis results.
               Performance tests conducted on both the original and the turbo PrintServer 20
               indicate the enhanced processing performance of the turbo controller.



In 1988 the turbo controller project was conceived          We were to use existing, qualified chips wherever
as a means of extending the l ife of the PrintServer 20     possible in order to avoid new part qualification
platform by introducing a performance-enhanced              costs and appl ication-specific integrated circui t
system controller. The system control ler in the            (ASIC) development costs.
PrintServer 20 is housed within and powered by                 Early investigations indicated that the p erfor­
the printer or "print engine"; it is a concise i mple­      mance target was indeed achievable with existing
mentation of a single-board computer containing a           inexpensive ruse processors, as well as a high­
CPU, a memory subsystem, an Ethernet interface,             speed Digital proprietary VAX processor. A ruse
and a printer interface. It suppl ies an environment        processor would require porting a 2 . 5 -megabyte
in which a multi tasking software system manages            (MB) software system, wh ich was far beyond the
communications with remote cl ients and with the            scope of the project. The highest performance
print engine, performs data conversion from the             VAX processor and the associated support chips,
page description language (PostScript) to bi t-map          which would not cause a problem with the soft­
images, and provides management of physical print           ware system, were far too expensive to be consid­
engine resources.                                           ered. Alternatives were therefore limited to less
   The original controller provided a maximum               expensive, lower speed VAX processors: the low­
print speed of 20 pages per minute, but this perfor­        risk, 60-nanosecond (ns) CMOS VAX or CVAX pro­
mance could not be maintained when the docu­                cessor was proven, ancl the higher speed and more
ment included complex text, graphics, or images. To         cost-effective "system on a chip " or SOC processor
i mprove page throughput for complex docu ments,            was u nder development. Ei ther choice would have
a controller was needed on which PostScript code            a minimal impact on the software system and
could execute faster. To enhance performance, the           would provide a cost-effective solution.
competi tion was moving toward control lers based              The original performance estimates for the CVAX
on new industry-standard reduced instruction set            and the soc processors in general-purpose process­
computer (RISC) processors. Therefore, to be com­           i ng environments were below the lower bound of
petitive, Digital's new controller was required to          the performance target. The design team was also
improve performance by five to eight times that of          uncertain of the actual execution characteristics of
the original control ler, which hacl been based on          the PrintServer software. For these reasons, it was
the rtVA.,'C m icroprocessor.                               decided to begin the project wi th a performance
   As challenging as the performance improve­               analysis of the origi n a l control ler to determine the
ment would be to achieve, budgetary pressures               expected performance of a design based on either
forced restrictions on the implementation strategy.         processor.



Digitttl Teclmicaljourmrl   Vol. 3 No. 4   Fall /991                                                             61
I mage Process ing, Video Te rminals, and Printer Technologies




     This paper discusses the problems encou ntered            The RETrACE hardware consists of the following:
t.l uring our analysis and the solutions d<:vised by the
                                                           •   Two ime rco nnect boards boot and operate a
Hart.lcopy Systems Engineering Group to overcome
                                                               system control ler on a table top. Developed as
them. The RETrACE tool suite, a performance ana ly­
                                                               part of the original PrintServer 20, the boards
sis system, is described and the analysis resu lts are
                                                               connect the controiJer to a prin t engine and an
provided. The paper then discusses the hart.lware
                                                               Ethernet.
architecture of the turbo control ler and ends with a
presentation of the performance resu lts obtai ned         •   The PrintServer 20 server controller was modi­
for standard PostScript benchmarks.                            fied for use as an intell igent trace bu ffe r system.

                                                           •   The PrintServer 20 server control ler's memory
Performance Analysis of the Original                           capacity (12MB) was extended using the standard
Controller                                                     4MB memory module used on the Kanji version
The Pri ntServcr 20 software system consists of a              of the PrintServer 20.
 VA-'\ ELN opera ting system, an Adobe Systems, Inc.
 PostScript interpreter, and a substantial amount of       •   The RETrAC E mother board was developed specif­
software to manage communications and resources.               ically for this tool su i te. I t contains a 32-bi t wide,
The task of analyzi ng its performance was compli­             first- i n , first-out (Fi fO) buffer a nd two loosely
cated by two additional factors. First, the software           coupled state machines.
system's behavior depended on the characteristics          •   A standard PrintServer 20 system controller and
of the user's PostScript document. PostScript is               print engine were used as the "system u nder
an interpreted program m ing language. Thus, like              observation."
any computer program, low-level machine perfor­
mance can be dramatically affected by the program          •   The console terminal was selected from the stan­
being executed . Second, ami more painfu l , the               dard VT series of term inals.
proprietary nature of the PostScript interpreter               A diagram of the RETrACE hardware system is
prohibited us from obtaining code sources, and d is­       shown in Figure 1.
cussing its i nternal architecture with engineers              The RETrACE mother board performed the data
from Adobe Systems.                                        capture, using the modi.fi ed control ler's memory as
    While the characterization of a complex, par­          a la rge bu ffer. The board m o n i tored the processor
tially proprietary, real -time software system is          bus of the system under observation by copying
difficult, it is not impossible. Programmer cou nter       al I addresses and com m u n ications between the
address (PC) traces have offered many systems              r tVA-'\ processor and its external floating-point
designers very detai led insight i nto the execution       unit. This copied data was placed into a FIFO bu ffer
performance and characteristics of systems. PC             that in turn was writ ten into the memory of the
traces provide a means to observe a system at a            mod ified controll er using a direct memory access
macroscopic level, allowing a view of the complex          (Oi'vlA) device. Since a standard PrimServcr 20 con­
i n teractions between the hardware a nd software          tro l ler and its optional memory expansion provide
systems. System designers can use captured address         16.VIB of storage, approximately 3 seconds of real­
trace; from current machine performance to extra­          time execution address traces cou ld be caprured.
polate expected performance of future systems and          The data capture continued until the trace bu ffer
help them make architectural trade-offs.                   memory was exhausted , at which point the data
                                                           was uploaded over a network connection to a VAX
The RETrACE Analysis System                                VMS computer for analysis.
The HETrACE tool suite was created to provide                  Due to the design of the original PrintServer 20
a nonintrusive means of capturing real-time PC             system, many large data areas and code sections
traces and ana lyzing the captured addresses. The          were mapped into different explicit memory spaces.
tool su ite consists of both hardware and software         This subdivision provided a means of determining
components.                                                which code function was executing in any given
   In order to keep expenses at a minimum, existing        segment of the add ress trace. With a simple statisti­
hardware was used wherever possible. Only one              cal study it was possible to generate software exe­
small module had to be developed to complete the           cution histograms and to determine many of the
RETrACE hardware platform.                                 characteristics of the system, including translation



62                                                                  Vr>l. 3 No. 4   Fall 1991   D igital Technical journal
                                                                        Design of the Turbo PrintServer 20 Controller



                                                             RETRACE
                                                             CONSOLE
                                                             TERMINAL




                                      PRINTSERVER 20
                                      SYSTEM
                                      CONTROLLER

                                      SYSTEM UNDER
                                      OBSERVATION


                                  Figure 1     RETrACE Analysis System Hardware


buffer, floating point, instruction stream (!-stream)             fou nd , then a m iss was signaled and the simula tor
versus data stream (D-stream), read versus write,                 would proceed to the next level of memory in the
and interrupt performance. Hit rates for fu l ly asso­            hierarchy.
ciative caches of separate !-stream and D-stream,                   Whenever a hit occurred at a given level, it
as wel l as a combined I- and D-stream cache, were                was logged and all levels of memory in the hier­
also provided. These hit rates were determined for                archy above it would al locate entries based on
first-level wri te- through caches from 128 bytes up              their defined al location rules. While this procedure
to 256 kilobytes (KB). Thus an upper bound for an                 indicated the memory system performance for
optimum-perform ance cached memory system                         a p roposed architecture, the overall system per­
was determined.                                                   formance was still u nknown . Using a simple rule
   Both processors under consideration possessed                  based o n the average execution time per address
the abi l i ty to access a memory subsystem at speeds             for the existing control ler, and scaling that time
greater than that achievable with existing low-cost               based on the clock speed increase of proposed pro­
dynamic random-access memory (DRAM) technol­                      cessors, an overa l l performance number was esti­
ogy. The performance nu mbers predicted by the                    ma ted for a system based on either processor with
processor groups indicated that cached memory                     any arbitrary memory architecture.
subsystems were required. Because these sub­
systems can be expensive and their performance is                 Benchmark Selection
subject to the pecul iari t ies of the software that              The RETrACE tools sui te provided the components
executes on them, a m u ltilevel memory simulator                 required to study the execution characteristics of
was developed to al low accurate studies to be per­               the PrintServer system without changing the char­
formed on proposed cache architectures.                           acteristics of i ts normal operation. The only diffi­
   The simulator was configured at ru n-time to sim­              culty was to narrow the focus of the benchmark l ist
ulate an arbitrary hierarchical memory system that                to provide a representative sample of PostScript
was N levels deep, with an arbitrary size, associa­               documents to print. Due to time constraints, the
tivity, performance, and behavior at each level.                  list was l i m i ted to five benchmarks.
The memory level nearest the processor was
defined as the first level, and the l as t as m a i n mem­        BM1   The BM 1 benchmark stresses those aspects of
ory. The simu lator processed a trace file by wa lk­              the system that convert the mathematical represen­
ing each address in the file through the memory                   tations of characters to bit-map representations,
hierarchy starting nearest the processor a t the first            wh ich comprise the form that is printed. This
level. If a copy of the address was found at a given              benchmark uses several fonts in standard character
memory level, then a hit was signaled and the next                orienta tions, stressing both very large and small
address was processed. If that address was not                    character sizes.



Digital Tecbnicaljour11al   Vol. 3 No. 4   Fall 1991                                                                63
Image Processing, Video Terminals, and Printer Tech nologies




R.\12   Of the same type as B.\1 1 , this benchmark           the overa l l h i t rate to approxim ately 81 to 84 per­
stn::sses the transforms from ma thematical to hit­           cent. As the ana lysis of the data progressed, i t was
mapped character representa tions; howcvt.:r, the             understood that the write data must be studied
characters printed are at arbitrary orie nt ations           very closely since it cou ld have a dramatic i mpact
with sizes ranging from typical to very small.               on the overal l cache miss rate. Du ring the cache
                                                              model simula tions, the hit rates of the 1-stream
BM3 The B�1.1 benchmark is one of the standard               were between 85 to 90 percent. However, the
benchmarks for PostScript performance qua l i fica­          D-stream h i t rates were between 35 to 45 percent,
tion. I t is a simpk: 41-page document that contains         with writes accou n ting for 60 to 90 percen t of the
st.: veral different fonts. The benchmark is designed        tota l D-stream misses. To achieve the greatest posi­
to characterize the standard text-hand l ing per­            tive effect on the h i t rate of the system, enhance­
formance of a printer. This benchmark is printed             ment of write-miss performance was the most
n.vice to ensure that a l l characters to be primed          advantageous. The two options to improve this per­
have been converted from ma thematical ou tl ines            formann: were either to implement a wri te-back
to bi t-map representations of the characters. Thus          cache or to add a write buffe r to the system. Further
the focus of the benchmark is to move the text               cache simu la tions showed that a write buffer wou ld
data through the system, to copy the character bit           provide an 8 to 16 percent overa l l system perfor­
maps to the 1 MB region in memory that conta ins             mance improve ment, which was equal to t h a t of a
the image to be printed , and to print the image. I t        write-back cache. The wri te bu ffe r, however, was
should b e noted that this is t h e only benchmark           the more stra igh tforward solution to implement.
that printed at engine speed on the Prin tScrver 20             Cache analys is revealed that the processors
system controller powered by the rtVAX system.               required different memory archi tectures. The CVA.X
                                                             had an internal l KB, two-way set associative cache.
HOUSE A binary image file, the HOUSE bench mark              This was to be configured as a mixed 1- and D-stream
was used to stress the com munications aspects of            cache. An addi tionai 32KH ro 64KB, two-cycle wri te­
the PrintServcr system.                                      through cache was to be added exte rnal ly. This
                                                             wou ld also be configured as a mixed I- and D-stream
SCHEll'! The SCHEM benchmark was a vector repre­             cache. A si ngl.e- longword, two-cycle wri te buffer
sentation of a logic schematic. This benchmark was           would provide enough buffering to reduce the
used to stress the PostScript interpreter's ability to       drama tic impact of write misses. The soc was
interpret nonnative PostScript code and to exh ibit          proposed to have an internal write-back cache
the characteristics of drawing vectors.                      between 5 KB and 8KB, with each l KB region mak­
                                                             ing up a single set. Cache simu l a tions indicated
Analysis Results                                             that with a minimum internal mixed 1- and 0-stream
The thrust of the analysis was to provide credible           cache of 5KB, five-way set associative, an externa l
evidence to support architectu ral and imp lemen­            data cache would have to be over 64KB to have even
tation trade-offs. The major areas of focus were             a negligible effect on ovnal l system performa nce.
                                                             Therefore no external cache was recom mended. To
•    Mem ory system organization
                                                             mi tigate the write-miss penalty, a two-cycle write
•    Printer in tnface performance                           buffer of 4 to 6 longwords was reco mmended.
                                                                As an accelera tion technique, the original
•    Main memory bandwidth
                                                             PrintServer 20 control ler contai ned a memory
•    Overa l l system performance                            access capability that a l lowed data written to mem­
                                                             ory to be logical.ly ORed with data that was al ready
Memory Syste-m Organization The statistical anal­            stored. This technique was particularly useful when
ysis of the trace information provided many clues            the software system was wri ting the image that was
to direct our investiga tion toward the optimum              u ltimately printed. As part of the process of gener­
mcmorv system archi tecture. The overa ll read-to­           ating an image to print, the individual characters
write ratio for the observed benchmarks ranged               appearing on a page must be copied from a region
from as low as 4 . 5 : 1 up to '5.'5: 1 , which me:1ns for   of memory ca l led the font cache to another region
a wri te-through cache system with a theoretical             cal led the frame bu ffe r. The frame buffer contains
100 percent read h i t rate, the writes would degrade        the actual data that is sent to the print engi ne.



64                                                                  Vol. 3 No. 4   Fall I')<)I   Digital Technical ]ourual
                                                                Design of the Turbo PrintServer 20 Controller



To complicate things, the data written to the frame          Overall System Performance The execution char­
buffer must be able to overlay data that may al ready        acteristics of the original PrintServer 20 provided
be there, thus requi ring a logical OR function.             some interesting surprises. Most floating-point
  When a document was printing at or near the                calculations were performed in double precision;
maxim u m engine speed of 20 pages per m inute,              and even more interesting, for each floating-point
analysis showed this low-level copying function              operation, there was a floating-point conversion
consumed approximately 20 p ercent of the total              from si ngle to double precision, and then back
system time al lotted to generate and print one page.        aga i n . Since the precise operations were not
Thus a logical OR fu nction i n the memory system            requ i red, a simple compi ler switch removed the
would reduce the number of memory data cycles                conversions and provided a 3 percent overa l l sys­
from "2 reads 1 write" to " 1 read 1 write," and             tem p erformance improvement for floating-point­
reduce the impact from a second read occupying a             intensive PostScript documents. A second su rprise
useful cache l ocation. Without this capabil i ty, the       came from the resu lts of the BM3 benchmark,
degrada tion would be between 5 and 10 percent of            which indicated a translation bu ffer hit rate of
overal l system performance when printing at or              85 percent. At the time of the discovery, the
near 20 pages per m inute. Therefore memory capa­            PrintServer 20 was configured with a standard
bil i ty with a logical OR function was recommended.         M ic roVA..'( processor ; however, by substituting an
                                                             rtVAX, which uses one less memory access to refer­
Printer interface Performance When a PrintServer             ence its page tables, an 1 1 percent system per­
20 is printing, every page that ex i ts the printer          formance improvement was achieved. With this
requi res the 1 MB frame bu ffer to be copied from           i mprovement,    the   rtVA.c'C   processor    provided
memory to the print engine interface. Changing a             enough power to al low the original PrintServer 20
program-controlled p ri n ter interface to one d riven       to ship with its 20-page-per-minute designation.
by a DMA device provided two significant advan­              This information led the turbo contro l ler designers
tages. The first was to reduce the rea l- time requ i re­    to determine that the transla tion buffer of the SOC
ments on the PrintServer software system, and the            would be large enough for a l l the entries requ ired .
second was to a l l ow for a l i m i tecl degree of paral­
lelism on the controller. The para l lel ism was due to      Results
the abil ity of the processor to co ntinue to execute
                                                             The final analysis revealed that the expected perfor­
from i ts cache memory system while the DMA
                                                             mance of a CVAX or SOC processor wou ld place
device accessed memory. The processor only stops
                                                             either design on the low side of the p e rformance
executing when a cache m iss occurs.
                                                             requi rement. Therefore close attention to detail
                                                             would be requ i red d ur ing the implementation
Main Memory Bandwidth With a CVAX processor
                                                             phase of the project as every ounce of performance
configured as recommended in the section Memory
                                                             m a t tered . The expectation was to have a choice
System Organization, the main memory system
                                                             between an SOC processor with a 40-ns cycle time
bandwidth requ irement of the processor was
                                                             and a CVAX processor with a 60-ns cycle time. The
60 percent. For the soc , it was 70 p ercent when
                                                             performance i mprovements of the two processors
an existing DRAl\1 contro l ler was used. A Dl\'lA­
                                                             are compared in Table 1 .
driven printer interface requi red 15 percent, and
an Ethernet interface required nom inal ly 4 percent
with bu rsts up to 20 p ercent. Each subsystem was           Table 1    Performa nce I m prove ment Relat ive
scrutinized to reduce its required memory band­                         to Original Pri ntServer 20 Contro l ler
width. The resulting recommendation was to add a
32-bit bus to the memory subsystem to provide a                                     soc                    CVAX
                                                             Be nchmark             Processor              Processor
dedica ted channel for a l l data being sent to the
printer interface. This provision would reduce               8M1                        4.7                   3.7
required memory bandwidth for the printer inter­             8M2                        4.9                   4.0
face from 15 percent to about 7 percent. The sys­            8M3                        4.3                   3.3
tem would then have a nominal memo1-y bandwidth              HOUSE                      4.9                   4.2
requirement of 71 percent for a CVAX system and
                                                             SCHEM                      4.7                   3.7
81 percent for an soc .




Digital Tecbtticaljournal   Vol. 3 No. 4   Fa/1 1')')1                                                              65
Image Processing, Video Te rminals, and Printer Technologies




   As the project schedule progressed, the risk asso­       In each case existing chips satisfied some of the
ciated with the new soc processor decreased. As          requirements for the subsystem. In the end these
this risk window col lapsed, i t was u nderstood that    chips met all ou r requirements, bu t only because
a turbo control ler based on the soc processor           tht.:y wt.:re used in ways not originally intt.:ndcd by
would not only perform better, but would also cost       the chip designers.
less as it would not require an external cache.
                                                         Main Memory
Turbo Controller Hardware Design
                                                         Since the SOC has a bus interface that is compati­
The turbo controller was destined for a relatively       ble with the CVAX chip, the most obvious chip to
high-end pri nter. Therefore the hardware archi tec­     use as a main memory control ler was the CVAX
ture had to provide maximum performance, even            memory control ler (CMCTL) chip.1 It responds to all
though this implementation would increase costs.         bus cycles generated by the soc, anct si nce it was
Based on the results obtained during RETrACE analy­
                                                         already used on a munber of pla tforms supported
sis, the hardware design had the fo ll owing imple­      by the VAXELN operating system, i ts use greatly
mentation goals:                                         simplified port ing VAX ELN to the turbo control ler.
•    The SOC would provide the CPU, the float ing­       However, the turbo controller requires two special
     point accelerator (FPA), and the cache subsystem.   mem ory modes that are not provided directly by
     No secon d-level cache would be implemented .       the CMCTL, namely OR mode and scan- erase mode.
•   A four- to six-entry write buffer would be           I t was essential to devise a way to include these
    implemented.                                         two modes if the CMCrL were to be used.
                                                             OR-mode memory is a technique used to improve
•    The transfer of bi t-map data to the print engine   performance during the wri ti ng of the page bit
     would requ ire a 32-bit DMA subsystem with scan­    map into memory (scan conversion). D uring nor­
     erase capabi l i ty.                                mal memory operation (cal led replace mode) , the
•    The memory subsystem would support OR-mode          dest i nation operand in memory is replaced by the
     memory access by the CPU and scan-erase access      sou rce operand . D u ring an OR-mode write cycle,
     by the DMA control ler.                             the destination operand is mod ified as fol lows:
   Although both the SOC and rtVAX chips comply          •   For each logicaJ zero in the source data being
with the VAX arch itecture standard and both are             writ ten, the corresponding dest ination bit in
conceptually very similar, they have significant dif­        memory rem ains unchanged .
ferences in the bus interface. For example, the
soc uses a quadword cycle (one 32-bit address fol­       •   For each logicaJ one in the source data being
lowed by two 32-bit data reads) to fil l one i nternal       written , the corresponding destination bi t in
cache block, while the rtV�'\ processor, which does          memory is written with the corresponding bit in
not support caching, does not usc this type of               the pattern register.
cycle. AJso, the clocking system on the SOC was
                                                         •   The p attern register is a 32-bi t register which
enhanced, and the timing rela tionships between
                                                             determines the "color" pattern of the "ink" being
signals were modified to improve performance.
                                                             written on the pagt.:.
  The changes to the SOC bus in terfact:, plus the
requ ired functional changes revealed by RETrACE            Figure 2 shows a portion of the logic between
analysis, meant that very l i ttle of the original       the CMCTL and the memory array that implements
PrintServer 20 con troller design cou ld be appl ied     the OR-mode fu nction in hardware. The OR-mode
to the new control ler. One of the first questions to    operation is accomplished by inverting the source
be answered before the design of the turbo con­          data and connecting it to 32 independent write
tro l ler cou ld begi n, was whether or not one or       enables of the memory array. When a zero is wri t­
more AS!Cs wou ld be requi red for the design. This      ten, it is inverted and the write cycle for that bit
question had to be answered for three subsystems:        becomes a read cycle, thus preventing any change
                                                         to the memory contents. When a one is written, it
•    Main memory
                                                         is inverted and the write is aiJowed to occur, but
•    Write buffer                                        the data actually wri tten depends on the value pre­
•    Bit-map data transfer subsystem                     viously wri t ten into the pattern register.




66                                                               vbl. 3 No. 4   Faii i')'JI   Digital TeciJIIicaljournal
                                                                    Design of the Turbo PrintServer 20 Controller



MEMORY DATA BUS FROM CMCTL                            32           Three operations occur during a single scan­
                                                                 erase cycle. Refer to the circuit drawing in Figure 3.
                                                                 1. The bus m aster asse rts the signal's "bit-map


          D�
                                          32
                        DATA
                        IN
                                 DATA
                                 OUT    �        [>        I--
                                                                   load" and "bit-map erase" and requests a masked
                                                                   write. The C1viCTL performs a read, and the bit
                                                                   map is read ontO the mem ory data bus.
         PATTERN                               READ              2. Bit -map data is automatica l ly transferred from
         REGISTER                              DATA PATH
                                                                    the memory data bus intO the FIFO bu ffer.
                                                                 3. The CMCTL performs a write. However, since
                                                                    the bit-map erase signal has disabled the data
  '---
          [> �          WRITE
                        ENABLE
                                                                    path a nd the pul l-down resistors have set the
                                                                    data-in l i nes to all zeros, the write cycle, which
                                                                    was intended by the designers of the chip as a
       OR-MODE          MAIN MEMORY                                 masked write, has in fact become a memory
       WRITE            ARRAY
       DATA PATH                                                    clear opera tion.

                                                                 Write Buffer
             Figure 2   OR-mode Circuit
                                                                 The LR3220 chip was chosen as the base for the
                                                                 write buffer subsystem. It provides a six-entry FIFO
   Two fea tu res of the CMCTL chip m ake it possible            buffer for address, data, and byte mask and detects
to im plement OR-mode memory. First, its 64MB                    whether the processor has requested a read at a
address space is divided i nto 4 arrays of 4 banks               memory location for which a write is stil l pending.
(16 banks total). Second, the CMCTL chip can selec­              I t also supports two opera ting modes: LR3000
tively disable parity checking on an array.                      mode and Harvard mode.
   The large address space of the CMCTL al lows the                  If it were not fo r the Harvard-mode fea ture, it
use of 2 arrays for replace mode and 2 arrays for                would have been more difficult to inclucle the
OR mode, since the turbo controller supports up to               LR3220 chip into the turbo control ler. The LR3000
32MB of memory. The control signals of the two                   processor, for which this chip was designed, has
sets of arrays are combined such that OR mode and                staggered address tim ing. Some of the address and
replace mode access the same physical memory,                    byte-mask bits are asserted on the fall ing edge of
though in diffe rent ways. Parity error detection                the clock, and the remaining bits are asserted on
is disabled on the OR-mode arrays; thus a read­                  the rising edge of the clock. When the LR3220 chip
through OR-mode address space cannot cause a par­                is configured in LR3000 mode, the processor sub­
ity error. This is necessary because OR-mode write               system must meet these timing requirements.
cycles may corrupt pari ty. Normal ly any bit map                However, when the LR3220 chip is configured i n
created using OR-mode write cycles is read using                 Harvard mode, a l l address, data, a n d byte-mask
OR-mode read cycles.                                             information is read at the same rising clock edge.
   The other special mode req uired for the main                     The basic strategy for including the wri te buffer
memory system is cal led scan-erase mode. It is an               into the turbo controller was to insert the write buf­
opera ting mode designed to i mprove bus util iza­               fer between the soc ancl the rest of the system as
tion during the transfer of the bit map from main                shown in Figure 4. The SOC wou ld issue read and
memory to a FIFO buffer connected to the printer                 write requests to the write bu ffer, ancl the write
data l i nes. This mode is made possible by a side               buffer would issue read and wri te requests to the
effect of the error-correcting code (ECC)/parity                 rest of the system. During CPU cycles the soc and
genera tion logic i n the CMCTL. Any time a masked               the write buffer have a master-slave rela tionsh ip
write occurs (any write other than an al igned long­             in which the SOC is the master. The rela tionship
word , such as a byte write), the destination long­              between the write buffe r and the rest of the system
word must fi rst be read by the CMCTL, then                      is also a master-slave relationship; however. the
combined with the bytes tO be wri t ten in order                 write bu ffer is the master. In fact, the wri te-bu ffer
to generate the parity or ECC check bits for that                output interface must look al most identical to the
longword .                                                       soc.




Digital Technicaljoumal    Vol. 3 No. 4   Pall 1991                                                                   67
Image Processing, Video Terminals, and Printer Technologies




                                                                       FIFO BUFFER




                                                                         [[IlJ
                                                                                                   DATA TO
                                                                                                   PRINT ENGINE




                    BIT-MAP LOAD                                              I
                     MEMORY DATA BUS FROM CMCTL                  32

                               RE PLACE-MODE
                               WRITE DATA PATH



                                    [>
                                                                                  32

                                                                                           [>
                                             32
                                                          DATA        DATA
                             '--
                                                                             r-+-                '--


                                                  i
                                                          IN          OUT




                                     I
                                                                                       READ
                    BIT-MAP ERASE                                                      DATA PATH



                    WRITE ENABLE
                    FROM CMCTL                      1     COMMON
                    ---
                      .::._
                        c__:_c__:_
                               c__:_______        -f� WRITE
                                                          ENABLE


                                                         MAIN M E MORY
                                                         ARRAY


                                         Figure 3       Scan-erase Circuit


  The structure of the write-buffer subsystem is              entries in the LR3220 chip have data, the bus cycle
shown i n Figure 5. The bus interface unit responds           generator (BCG) removes the next entry ancl issues
to read or write requests from the soc. During                a write request to the appropriate subsystem.
write cycles, the bus in terface writes the data into            The write-buffer subsystem al lows the SOC to
the LR3220 chip and im mediately alerts the SOC to            " read around" the write buffer, provided rhe address
terminate the cycle quickly. Whenever om� or more             being read does not have a pending write in the


                                                                                                       TO OTHER
                      SYSTEM               WRITE-BUFFER                                                SUBSYSTEMS
                      ON A CHIP     1---   SUBSYSTEM



                                                           CVAX                            UN IVERSAL
                                                           MEMORY                          DI RECT MEMORY
                                                           CONTROLLER                      ACCESS
                                                           (CMCTL)                         CONTROLLER


                                                    I                                  I
                                              PRINTER                             MAIN
                                              DATA FIFO                           MEMORY
                                              BUFFER                              ARRAY


                                                    I
                                           TO PRINT ENGINE


                        Figure 4    Interconnection of Turbo Controller Subsystems


68                                                                       Vol. 3 No. 4       l'i:t/1 1')91   Digital Tech11ical jourt�al
                                                                           Design of the Turbo PrintServer 20 Controller




                                                                 LOCAL         BUS CYCLE
                                               BUS                             GENERATOR                      SYSTEM CONTROL
    SOC CONTROL                                INTERFACE         CONTROL
                                                                 SIGNALS       AND
                                               UNIT                            ARBITRATOR
                                                            I--          r--
                         r--
                                        ADDRESS IN
                       r-I>
                                                                                               r---.
    SOC DATA/ADDRESS
                         ...._
                         LATCH
                                           DATA I N
                                                           L_                  ADDRESS OUT
                                                                                 DATA OUT
                                                                                                          SYSTEM DATA/ADDRESS


    SOC BYTE MASK
                                                                                               v
                                                                                            MULTIPLEXER
                                                           LR3220 WRITE BUFFER                               SYSTEM BYTE MASK


                                                      Figure 5    Write Buffer


LR3220. To handle this, the BCG includes an arbitra­                 Bit-map Transfer Subsystem
tion circuit. When the SOC requests a read cycle,                    The bit-map transfer subsystem transfers bit-map
the bus interface unit of the write bu ffer passes                   data, created by the PostScript interpreter, to the
the request to the BCG. The BCG responds once                        print engine. It is composed of the 32-bit DMA con­
it has completed any write cycle cu rren tly in                      trol ler, a FIFO subsystem, and scan-erase logic in
progress, provided that the address to be read does                  main memory as described in the section Main
not have a pending write in the write buffer. When                   Memory.
the slave device being read acknowledges the BCG ,                     The main requ irements for the 32-bit DMA con­
the acknowledgment is passed back to the bus                         troller were
interface ancl finally to the soc to terminate the
                                                                     •   32 MB address range
cycle. The BCG then resumes its task of removing
entries from the LR3220 chip and issuing writes to                   •   Abil ity to transfer 32 bits at a time
the rest of the system.                                              •   Abil ity to transfer the frame buffe r forward
   In order to maintain data coherency, the wri te­                      (incrementing the source add ress) or backward
buffer subsystem enforces some additional                                (decrementing the source address)
protocols.
                                                                        None of the available DMA contro l ler chips met
•    Al l wri tes to any location other than main                    all our requirements, but the AMD 9516 universal
     memory require a wri te-fl ush cycle; that is, the              DMA controller (UDC) met some of them. The uoc
     bus interface mu st wait until the LR3220 chip                  is a 16-bit DMA controller with a 16MB address
     is empty before wri ting the data to it. Further­               range and the ability to increment or decrement the
     more, the bus inte rface must wait u ntil the BCG               source address. There were two drawbacks to the
     has finished the cycle before it acknowledges the               use of this chip. The software would have to ensure
     SOC and al lows it to pe rform the next cycle.                  that the frame buffer was always within the lower
                                                                     16MB of memory, and the UDC would use twice as
•    All reads to any location other than main mem­
                                                                     much bus bandwidth since it could transfer only
     ory require a read tl ush, which has the same
                                                                     16 bits at a time.
     restrictions as a write fl ush. These restrictions
                                                                        It was proposed that the UDC cou ld be used as a
     are required to avoid the possibil i ty of reading
                                                                     ful l 32-bit DMA controller if i t was connected to the
     around a pendi ng 110 space write, which often
                                                                     bus "incorrectly" by shifting the data/address l i nes
     has side effects to other addresses.
                                                                     to t he left by one bit. That is, data/address l ine 0 on
•    The write-buffer subsystem must pass a l l DMA                  the UDC would be connected to data/add ress line 1
     bus transactions to the soc to ensure that a l l                on the bus; data/address line 1 of the UDC would be
     cached memory locat ions that are modified by                   connected to data/address line 2 on the bus; etc.
     DMA cycles have their correspond ing cache                      This type of connection doubles the address range
     entry inva lidated.                                             of the chip and causes the sou rce address on the



Digital Technical journal     14">1. 3 Nu. 4   Fa/1 1991                                                                    69
Image Processing, Video Terminals, and Printer Technologies




bus to i ncrement by 4 bytes (32 bits) instead of                  Table 3           Benc h m a r k F i le Attributes
2 bytes (16 bits).
                                                                   F i le N a m e               General Attributes o f F i l e
     This decision had a few impl<.:mcntation impacts.
For example, the register definitions were now                     B M 1 . PS                   Contains 39 pages of text with
incorrect, since a l l the bits in a l l the registers were                                     13 fonts of various si zes. Some
shifted one b i t to the left. However, once the soft­                                          text strings are at varyi ng angles.

ware was modified to compensate for this, the UDC                  B M 2 . PS                   Contains 1 page of spiral text
functioned properly as a 32-bit DMA control ler.                                                of va rious point sizes.

When combined with the scan-erase feature of                       BM3.PS                       Contains 41 pages of text with
                                                                                                5 fonts.
main memory, i t a llowed u s to achieve our b i t -map
                                                                   HOUSE.PS                     Contains a 1 -page bitonal image
transfe r goal of reading 32 hits from memory, load­
                                                                                                of 3000 blocks (DECnet l i m ited).
ing i t into the FII'O subsystem, and clearing the
                                                                   SCH E M . PS                 Contains a 65-page schematic
memory location, all in a single DMA cycle.
                                                                                                of g raphics (vecto rs) and text.

Pe rformance
In this section, the performance of the original                   Math Operators Performance of the
PrintServer 20 is compared to the enhanced perfor­                 PostScriptjob
mance of the turbo PrintServer 20.                                 Figure 6 i l l ustrates the control lers' performance
     Except for performance, the original PrintServer              resu lts in math operations per second. The test
20 and the turbo PrintServer 20 have identical func­               determines the time needed to perform 50,000
tional capabi l i ties. Table 2 l ists the five functional         primitive m a th operators (e.g. , adding two mJm­
subsets that were characterized for performance                    bers 50,000 times) during a PostScript test docu­
on both printers. The first fou r fu nctional subsets              ment. The real-time operator reads the current
were rated using the PostScript real-time operator;                time, and the repeat construct repeats the math
they measure the elapsed CPU time needed to                        operator. This test measures the performance of
complete a test. The last functional subset was                    the CPU only.
rated according to the          rate of pages exi ting
the printer. The term " DECnet/DPS" refers to the
                                                                   0 1 4000
DECnet job (a job is one of several m u ltiprocessing
                                                                   z
tasks running on the controller) and the '"dis­                    8 1 2000
                                                                   w
tributed    PrintServer software''         job.       The   term   (/)
                                                                   a: 1 0000
" printer system" refers to the complete printer                   w
                                                                   Cl..
system, i ncluding the PostScript job and the p r int­             (/)   8000
                                                                   z
ing overhead jobs. The printer system was rated                    0
accord ing to the rate of pages exiting the printer.
                                                                   �     6000
                                                                   II
  Table 3 reports the general attributes of the five               �

                                                                   � :
                                                                         4000
                                                                   0
files that were run with the RETrACE system and                    I

                                                                       I I _[ I I I I
                                                                         200
chara<:terized for performance.

                                                                                   ADD      DIV     MUL     SORT   COS     EXP   LOG

Ta ble 2     Functional Subsets of the Pri nters                   KEY:

                                                                   0 OR I G I NAL CONTROLLER
Functional Subsets        C h a ra cter iza t i o n                • TURBO CONTROLLER
                                                                   T U R BO = 6.7 x O R I G INAL CONTROLLER
PostScript job            Math operations per second

PostScript job            Text: characters per second

PostScript job            G raphics: vector inches per
                                                                    Figure 6          PostScriptjob Performance with Math
                          second

DECnet/DPS jobs           DECnet/DPS: kilobytes per                Text Performance of the PostScriptjob
                          second                                   Figure 7 compares the text performance of the
Printer system            I mage pri n ti n g : square             PostScript job on the original controller and the
exc l u d ing             inc hes per second
                                                                   turbo control ler. The test determines bow long it
DECnet/DPS
                                                                   takes the PostScript job to compose 250,000 equally



70                                                                              14>1. 3 No. 4   Fall 1991   Digital Technical journal
                                                                         Design of the Turbo PrintServer 20 Con troller



      20000                                                     Image Performance
                                                                The i mage test characterized the complete printer
0
z
0
                                                                system, inclu d i ng the PostScript job and the print­
()    1 5000
w                                                               ing overhead jobs, but excl ud ing the DECnet/ DPS
(f)
a:                                                              time requi red to transfer an image file to a printer.
w
(l_
(f)   1 0000                                                    Three one-square-inch bitonal i mages at device
a:
w                                                               resolution were placed into the user dictionary
f-
()
<>:
                                                                and were used repeated ly d uring the performance
a:     5000
<>:                                                             measurement. The result of using these precached
I



                 I I I, _I, I I
()                                                              i mages was to el iminate the DECnet and DPS soft­
                                                                ware t ime that would be requi red to transfer a full­
             0
                  2      4                         10      12   page i m age from a host to the printer. Performance
                             FONT P O I N T SIZE
                                                                was measured by printing 10 pages of 80 square
KEY:                                                            inches of image per page.
0 O R I G I NAL CONTROLLER                                              The pages were printed landscape and portrai t
•     TURBO CONTROLLER
                                                                t o measure the i mage performance both o n axis and
TURBO = 5 . 2 x O R I G INAL CONTROLLER
                                                                off axis. (On axis means that the printer sequen­
  Figure 7        PostScriptjob Performance with Text           t ia l ly prints a l l b i ts of a word from the image on a
                                                                single scan line. Off axis by 90 degrees means that
sized characters to the page buffer i n memory,                 the printer prints one bit from each word and does
which eventually is sent to the print engine to be              not print the next bit i n the word until it is a t
printed.                                                        t h e same position o n t h e next scan l ine.) Figure 9
                                                                shows the results of the image p erformance test in
Graphics Performance of PostScriptjob                           square inches per second .

An important means of characterizing graphics per­
formance is in vector inches p e r second. Figure 8                     22.0

shows the results obtained by running a PostScript              0       20.0
                                                                z
vector program in which all vectors are at                      0
                                                                ()
                                                                        1 8.0
                                                                w
45 degrees and vector lengths are from 0.1 inch to              (f)     1 6.0
                                                                a:
3 inches.                                                       w       1 4 .0
                                                                (l_
                                                                (f)     1 2.0
                                                                w
                                                                I       1 0 .0
                                                                ()
      3000
                                                                ?;       8.0
                                                                w
0                                                               a:       6.0
15 25oo                                                         <>:
                                                                ::::J    4.0
()                                                              a
w                                                               (f)
                                                                         2.0
� 2000
w                                                                           0
(l_
� 1 500
I                                                               KEY:
()
?; 1 000                                                        0 O R I G INAL CONTROLLER
a:                                                              • TURBO CONTROLLER
0
f-                                                              T U R BO = 3.0 x O R I G I NAL CONTROLLER
()    500
w
>


                                                                        Figure 9   Image Performance Measurement
                                                                                   of the Printing System
KEY:

0 O R I G I NAL CONTROLLER                                      DECnet/DPSjobs Performance
• T U R BO CONTROLLER
                                                                DECnet/DPS transfer rates can be ignored for text
TU RBO = 4 . 6 x O R I G I NAL CONTROLLER
                                                                and graphics files, but these rates can consume
                                                                most of the time needed to print large image files.
             Figure 8   PostScriptjob Performance               For example, a single, letter-size page of image
                        with Graphics                           contains more than 1 MB of i mage data, but the



Digital Technical journal       Vol. j No. 4   Fall 1991                                                                71
Image Processing, Video Terminals, and Printer Technologies




corresponding PostScript file con tains morT t han             to file printed and by the amount of CPU time used
2 M B . Because the image data is represented in Amer­         to print the job. For example, in t he B.\1 :) bench­
ican standard code for information interchange                 mark, the speed is limited by the 20-page­
(ASCII) hexadecimal characters in PostScript, S bits           per-minu te print engine, but the CPU time needed
of the PostScript file are needed to represent 4 hits          to print the file can be used as a performance
of image data.                                                 measurement.
   'I(> measure DECnct/ DPS, a PostScript file of I M R
o f comments was sent t o the printer. The clock was          Summary
started ·when the beginning of the file was tTccived          The turbo controller enhanced the performance of
by the PostScript interpreter and stopped when the             the PrintScrver 20 printer system. Its design was
end of the file was received . The assumption of this         prompted by the need to maintain print speed
test method was that the PostScript interpreter can           performance for complex documents containing
parse comment I ines much faster than DECnet/ DPS             text, graphics, and images. The RETrACE system was
can t ransfer them.                                           designed to analyze the PrintServer 20 system to
   The DECnet/DPS transfer rate is basica l ly propor­        determine which architectural changes would pro­
tional to the slower of the host and printer proces­          vide the greatest improvement in PostScript perfor­
sors. Figure 10 shows the DECnct/DPS results.                 mance. By optimizing hardware only in areas where
                                                              it was truly worthw hile, we were able to use exist­
RETrACE Benchmark Files                                       ing chips and reduce development costs. The sub­
The benchmark files l isted in Table 4 arc charac­            systems of the turbo control ler hardware that
terized both by the elapsed time from file arrival            were optim ized as a resu lt of this analysis were
                                                              the processor (SOC which provided CPU , floating­
                                                              point accelerator, and cache subsystem), a memory
       90
                                                              subsystem with OR-mode and scan-erase access,
       80
0
                                                              a write-bu ffer subsystem, and a 32-bit DMA sub­
z      70                                                     system. Results of the p erformance tests for five
0
()
w      60                                                     benchmarks, includ ing PostScript jobs, indicate the
(/)
cc     50                                                     levels of enhanced performance.
w
a_
(/)    40
w                                                             Acknawledgments
f--
>-     30
C)
0                                                             Chris Mayer designed and i mplemented the
...J   20
52                                                            RETrACE multilevel cache simulator. He developed
       10                                                     a ticketing algorithm that simpl ified the manage­
        0                                                     ment of delayed behavior memory constructs such
KEY:                                                          as write buffers.
0 O R I G I N A L CONTROLLER
• TURBO CONTROLLER                                            Refe rence
TURBO = 4.9 " O R I G I N A L CONTROLLER
                                                              1 . D. K. Morgan,       "The CVAX CMCTL-A CMOS
                                                                    Memory Control ler Chip," Digital Technical
       Figure 10      DECnet/DPS}obs Perjonnance                   journal, vol. 1, no. 7 (August 1988): 139 - 143.



Table 4      Bench m a r k F i les Cha racterized by Ela psed Time a n d CPU Ti m e (Seconds)

Be nchmark               Original          Tu rbo         Original           Tu rbo                  delta          delta
File                     CPU               CPU            Elapsed            Elapsed                 CPU            Elapsed


BM1                        7585            1 707           7735                2050                   4.4             3.8

8M2                          238             51             241                   51                  4.7             4.7

8M3                           56              15            1 28                 1 20                 3.7             1 .1 *

HOUSE                         67              15            1 06                  31                  4.5             3.4

SCHEM                      2802             625            3073                  675                  4.5             4.6
"Limited by engine.




72                                                                     Vol. 3 No. i     Full 1991   Digital Technical journal
              I   Further Readings


                  The Digital Technical Journal            DECwindows Program
                  publishes papers that explore            Vol. 2, No. 3. Summer 1990
                  the technologicalfoundations             An overview and descriptions of the enhance­
                  of Digital's major products. Each        ments Digital's engineers have made to MIT's
                  J ournal fixuses on at least one         X Window System in such areas as the server, tool­
                  product area and presen ts a             kit, interface language, and graphics, as we l l as
                  cornpilation of papers written           contributions made to related industry standards
                  by the engineers who developed
                                                           VAX 6000 Model 400 System
                  the product. The contentfor
                                                           Vol. 2, No. 2, Spring 1990
                  the Journal is selected by the
                                                           The h ighly expandable and con.figurable midrange
                  journal Advisory Board.
                                                           fami l y of VAX systems that includes a vector proc­
                  Digital engineers who would
                                                           essor, a high-performance scalar processor, and
                  like to contribute a paper
                                                           advances in chip design and physical technology
                  to the Journal shou/.d contact
                  the editor at RDVAX.::BLAKE.             Compound Document Architecture
                                                           Vol. 2, No. 1, Winter 1990
Topics covered in previous issues of the Digital           The CDA family of archi tectures and services that
Technicaljournal are as fol lows:                          support the creation, interchange, and processing
                                                           of compound documents in a heterogeneous net­
Availability in VAXcluster Systems/                        work environment
Network Performance and Adapters
                                                           Distributed Systems
Vol. 3, No. 3, Summer 1991
                                                           Vol. 1, No. 9,june 1989
Discussions of VMS volume shadowing, VAXcluster
                                                           Products that a l l ow system resource sharing
application design, and new ava i labil i ty features of
                                                           throughout a network, the methods and tools
local area VAXcluster systems, together with detai ls
                                                           to evaluate product and system performance
of high-performance Ethernet and FDDI adapters,
and an analysis of FDDI LA N performance                   Storage Technology
                                                           Vol. 1, No. 8, February 1989
Fiber Distributed Data Interface                           Engineering technologies used in the design, man­
Vol. 3, No. 2, Spring 1991                                 u facture, and maintenance of Digital 's storage and
The FDDI IAN system and Digital's products that            information management products
support this technology, with an overview and
papers on the physical and data l i nk layers,             CVAX-based Systems
Common Node Software, bridge and concentrator              Vol. 1, No. 7, A ugust 1988
devices and related management software, and an            CVAX chip set design and multiprocessing archi­
ULTR.IX network adapter                                    tecture of the m idrange VA.,'\ 6200 fam i l y of systems
                                                           and the MicroVA.,'< 3500/3600 systems
Transaction Processing, Databases, and
                                                           Software Productivity Tools
Fault-tolerant Systems
Vol. 3, No. 1, Winter 1991                                 Vol. 1, No. 6, February 1988
                                                           Tools that assist programmers in the development
The archi tecture and products of Digital's d is­
tribu ted transaction processing systems, with             of high-quali ty, rel iable software
information on moni tors, performance measure­             VAXcluster Systems
ment, system sizing, database availabi l ity, commit       Vol. 1, No. 5, September 1987
processing, and fau l t tolerance                          System com m unication architecture, design and
                                                           implementation of a d istributed lock manager, and
VAX 9000 Series
                                                           performance measurements
Vol. 2, No. 4, Fall 1990
The technologies ami processes used to build               VAX 8800 Family
Digi tal's first mainframe computer, including             Vol. 1, No. 4, February 1987
papers on the architecture, m icroarchi tecture,           The microarchitecture, internal boxes, VAXBI bus,
chip set, vector processor, and power system,              and VMS support for the VAX 8800 high-end mul ti­
as wel l as CAD and test methodologies                     processor, simulation, ancl CAD methodology



Digital Tecbnical ]ou mal    Vol. 3 No. 4   Fall /')91                                                           73
 Further Readings



 Networking Products                                      J Delahu nty and T. Kielty, "Automated Pareto
 Vol. 1, No. 3, September 1986                            Analysis for Continuously Improving a VLSI
 The Digital Network Architecture (DNA), network          Fabrication Area's Process Stabi l i ty," Advanced
 performance, LANbridge 100, DECnet-l i LTRIX and         Semiconductor A1anufacturing Conference
 DECnet-DOS, monitor design                               (September 1990).
 MicroVAX II System                                       S. Del l , " Promoting Equal ity of the Sexes through
 lf)/. 1, No. 2, March 1986                               Technical Writing," Society. for Technical Commu­
The implementation of the microprocessor and              nication (August 1990).
floating point chips, CAD su ite, MicroVAX work­
                                                          B . Doyll' and K. Mistry, "A Lifetime Prediction
station, disk controllers, and TK';O tape drive
                                                          Method for Hot-carrier Degradation in Surface­
VAX 8600 Processor                                        channel P-MOS Devices," IEI:.'F [i-ansactions on
 Vol. 1, No. 1, August 1985                               Electron Devices (May 1990).
The system design with pipel inecl architecture,
the 1-box, F-box, packaging considerations, signal        E. Freedman and Z. Cvctanovic, " Efficient
integrity, and design for reliability                     Decomposi tion and Performance of Para l lel
                                                          POE, FFT, Monte Carlo Simulations Simplex
Subscriptions to the Digital 1'echnical.fournal are
                                                          ami Sparse Solvers," IEEE Supercomputing '90
available on a yearly, prepaid basis. The subscrip­
                                                          (November 1990).
tion rate is 540.00 per year (four issues). Requests
should be sent to Cathy Phi l l ips, D igital Equipment   A. Garde! and P Deosthali, " Hub-centered Pro­
Corporation, M LO I-3/B68, 146 Main Street, Maynard,      d uction Control of Wafer Fabrication," Advanced
MA 01754, U.S.A. Subscriptions must be paid in U.S.       Semiconductor Manufacturing Conference
dol lars, and checks should be made payable to            (September 1990).
Digital Equipment Corporation.
                                                          A. Hartzel l , " I ntroduction of Argon as a Heat
Single copies and past issues of the Digital              Transfer Gas in a Single Wafer RJ E System ,"
Technical.fournal can be ordered from D igital            Jntenwtional Symposium for Testing and
Press at a cost of $16.00 per copy.                       Failure A nalysis (October 1990).

Technical Papers by Digital Authors                       A. Heyman and J Thottuvelil, " L inear Averaged and
                                                          Sampled Data Models for Large Signal Control of
R. Al-Jarr, ''A Methodology for Evaluating Decision       High Power Factor AC-DC Converters," IEEE Power
Making Architectmes for Automated Manufactu r­            Electronics Specialists Oune 1990).
ing Systems,'' Eleventh JF-AC Conference (August
1990).                                                    L. Hill, "Video Signal Analysis for EMI Control,"
                                                          IEEE Electromag '91 (1991).
S. Angebranndt, R. Hyde, D Luong, and N. Siravara,
"Integrating Audio and Tel<.:phony in a Distribu ted      L. H i l l and A . Metsler, "Video Subsystem Design

Workstation Envi ronment," Proceedings of the             for EMI Control," IEEE Electrmnag '91 (1991) .
Summer 1991 T!SLNIX Conference Oune 1991 ).               S . Kasturi, " Forced Convection: The Key t o the
                                                          Versatile Reflow Process," NEPCON East '90
S . Batra, M . Mallary, a n d A. Torabi, " Frequency
                                                          Oune 1990)
Response of Thin-lilm Heads with Longitudinal
and Transverse Anisotropy," !J:'I:'f:' lntermag '90       D. Mirchandani and P Biswas, " Characterization
(April 1990).                                             and Modeling Ethernet Performance of
                                                          Distributed DECwinclows Applications,"
R. Csencsits, N. Riel, ]. D ion and S. Arsenault,
                                                          ACM Sigmetrics (May 1990)
"Interfacial Structure and Adhesion of Metal-on­
polyamide," International Symposium for Testing           W Metz, "Au tomated On- l ine Optimization of an
and Failure Analysis (October 1990).                      Epi taxial Process," International Semiconductor
                                                          Manufacturing Science Symposium (May 1990).
R. Csencsits, J. Rose, R. St. Amand, L. E ll iott,
A. Hartzel l , L. Kisselgof, and ]. Lloyd, "Aluminum      K. Mistry, B. Doyle, and D. Krakauer, "Im ract of
Interconnect Microstructure and Its Role in               Snapback I nduced Hole Injection on Gate Oxide
Electromigration," Jntentational SymjJOsium               Rel iabil ity in N-MOSFET's," //.;/:1:.' Electron Device
for Testing and Failure A na�ysis (October 1990).         Letters (October 1990).



74                                                                Vol. .) No. .:j   Fall I'J'JI   Digital Technical journal
                                                                                                                 I
C. Pan, "Gas Lubrication," ASME/STLE Tribology          VAX/VMS: Writing Real Programs in DCL
Conference (October 1990).                              Paul C. Anagnostopou los, 1989, softbou nd,
                                                        409 pages, Order No. EY- C 168 E -D P-EEB ($ 29.95)
A. Philipossian, "Fluid Dynamics Analysis of
Thermal Oxidation Systems via Residence Time            This book contains information that can help the
Distribution (ROT)," Electromechanical Society          reader Jearn to write powerful and wel l-organized
(October 1990).                                         programs in DCL, the command language for the
                                                        VAX/VMS operat i ng system. The text includes a
M. Sidman, "Convergence Properties of an
                                                        review of the syntax and semantics of DCL and a
Adaptive Runout Correction System," ASME
                                                        discussion of significant issues in the development
Winter Meeting (November 1990).                         of serious DCL software. Programming paradigms
M. Sidman, " Parametic System Identification            are presented, as wel l as the correct way to
on Logarithmic Frequency Response Data,"                implement them. The book presents good pro­
IEEE Transactions on Automatic Control                  gramm i ng techniques and helps the student to
(September 1991 ).                                      make effective use of the VMS operating system .

D. Skendzic, "Two Transistor Flyback Converter
                                                        X WINDOW SYSTEM TOOLKIT:
Design for EM! Control," IEEE Symposium on
                                                        The Complete Programmer's Guide
Electromagnetic Compatibility (August 1990).
                                                        and Specification
A. Smith and W Gol ler, " New Domain Configura­         Paul ). Asente and Ralph R. Swick, 1990, softbound,
tion in Thin-film Heads," Intermag '90 (April 1990).    1000 pages, Order No. EY-E757E-DP-EEB ($44.95)

H. Smith and ). Beagle, " SIMS for Accurate             This book consists of two parts, "Programmer's
Process Monitoring i n CoSi2-on-Si MOSFET               G uide" and "Specification." "Programmer's G u ide"
Technology," Secondary Jon Mass Spectrometry            describes how to use the X Toolkit to write
(September 1989).                                       appl ications and widgets, and i nc ludes many
                                                        examples. Each chapter i n this part contains an
). Thottuve l i l , "Using SPICE to Model the Dynamic
                                                        appl ication writer's section and a widget writer's
Behavior of OC-to-DC Converters Employing Mag­
                                                        section. Application program mers need to read the
netic Amplifiers," IEEE Applied Power Electronics
                                                        widget writer's sections only if they are curious
Conference (March 1990).
                                                        about what is going on behind the scenes;
R. Ulichney, "Frequency Analysis of Ordered             widget program mers shou ld read both sections.
Dither," Hard-copy Output OE!LASE 89 SPIE '89           " Specification" provides a complete and concise
Proceedings (1991).                                     description of every component of the X Too l k i t
                                                        Intrinsics, a s standardized b y t h e MIT X Consor­
R. Ulichney, "Challenges in Device Independent
                                                        tium. The level of detail i n this part is sufficient
Image Rendering," Applied Vision Optical Society
                                                        to enable a program mer to create a new imple­
ofA merica Technical Digest Series '89 (1991).
                                                        mentation of the X Toolkit.
E. Zimran, " Performance Efficient Mapping of
Applications to Paral lel and Distri buted Archi­       PRODUCTION SOFTWARE TIIAT WORKS:
tectures," International Conference on Parallel         A Guide to the Concurrent Development
Processing (August 1990).                               of Realtime Manufacturing Systems
                                                        john A. Behuniak, Iftikhar Ahmad, and
Digital Press
                                                        Ann l\1 . Courtright, 1992, softbound , 204 pages,
D igital Press is the book publ ishing group of         Order No. EY-H895E-DP-EEB ($24.95)
D igital Equipment Corporation. The Press is an
                                                        This is a practical gu idebook for manufacturing
i nternational publ isher of computer books and
                                                        m anagers and process engineers who must develop
journals on new technologies and products for
                                                        better process methodologies to stay competitive
users, system and network managers, program­
                                                        and for developers of realtime manufacturing
mers, and other p rofessionals. Proposals and ideas
                                                        software who need to cut time and costs from their
for books in these and related areas are welcomed .
                                                        work. The presentation, which provides useful
The fol lowing book descriptions represent a            advice ancl easy-to-follow procedures, addresses
sample of the books available from Digital Press.       three basic tasks of realtime software development



Digital Technical journal   Vol. 3 No. 4   Full 1991                                                            75
Further Readings



in a manufacturing plant: ( 1 ) managing the design          VAX Pascal progra m m ing language, this is the first
of the system; (2) setting up and managing a                 book to actual ly discuss the construction of real
development organization; and (3) implementing               VMS applications. It sets forth a methodology for
tools for successfu! completion and managem<..: nt.          producing h igh-qua l ity, p rofessional VMS appli­
                                                             ca tions by focusing on the aspects of the V M S
UNIX FOR VMS USERS                                           opera ting system crucial to every well-written
Phi l ip E. Bourne, 1990, softbound, 36R pages,              application.
Order No. EY-C 177E-DP-EEU (S28.95)

This book emphasizes the practical aspects of                THE DIGITAL GUIDE TO SOFTWARE
making the transition f rom the ViviS to the U N I X         DEVELOPMENT
operating system. Every concept p r<..: s ented i s          The staff of the Corporate User Information
il lustrated w i t h o n e or more examples, comparing       Products (CUIP/ASG), D igital Equ ipment
how to perform a par ticular task in each of the             Corpora tion, 1989, softbound, 239 pages,
two operating systems. The book is organized in a            Order No. EY-C l 78E-DP-EEB ($ 27.95)
logical order and covers the following topics: fun­          THI:' DIGITA L GUIDE TO SOF1 mlRE DEVELOPNIENT
damental concepts to be grasped before touching              is the tirst publ ished description of the method­
the keyboard, the first terminal sessions, the first         ol ogy that Digi tal uses to design and develop i ts
commands, editing, com mu nica ting with users,              software . For the engineer ancl other professionals
resource u t i l iza tion, using devices, more advanced      associated with the creation ami marketing of
commands, using h igh- level languages, program­             software applications, this book gives a rare look a t
ming the operating system, text processing, and              t h e practices o f a n industry leader and provides a
networking. Appendixes provide extensive cross­              model for others who wish to introduce software
reference tables to make this a valuable reference           enginet:ring methods and tools i nto their own
tool for even the experienced UNIX user.                     companies. Also discussed are the use of selected
                                                             VMS case tools to exped i te the process; the roles of
LOGISTICAL EXCELLENCE:
                                                             teams and team leaders; the use of review meetings
It's Not Business as Usual                                   and docu ments; and formal procedures for testing
Donald ). Bowersox, Pa tricia J Daugherty,
                                                             and ma intenance. The guide includes numerous
Cornel ia L. Drogue, Richard N. Germain, and                 diagrams and tables, c l ear guidel ines for the coding
Dale S. Rodgers, 1992, 300 pages,                            and documentation of software modules, a l isting
Order No. EY-1 1953E-DP-EEB                                  of related VMS documen tation, and coding gu ide­
This book focuses on the interpretation of research          l ines for VAX C.
findings that have been compiled to help managers
who seek to improve logistical competency within             DIGITAL GUIDE TO DEVELOPING
their organization . It provides a sequent ia l model,       I NTERNATIONAL SOFTWARE
the best p ractices of "excellent" logistics managers        The staff of the Corporate User Information
with supportive statistica l evidence, and extensive         Products (O J I P/AS< i), D igita l Equipment
coverage of Electronic Data Interchange in the               Corporation, 1991, softbound, 381 pages,
logistics process. It also includes a brief overview         Order No. EY-F577E-DP-EEB ($28 95)
of the expanding role that logistics has recently
                                                             This book introduces the grou nd-breaking pack­
played i n the overall corporate strategy of increas­
                                                             aging and design gu idelines recommended by
ing speed and qual ity. lb facili ta te i nterest and ease
                                                             Digi tal for products destined for overseas markets.
of reading, an action-oriented case d ia logue ru ns
                                                             Al ready used by more..: than 400 independent soft­
throughou t the eight chapters.
                                                             ware vendors and development groups, as wel l as
                                                             by Digital engineers, this book offers an approach
WRITING VAX/VMS APPLICATIONS USING
                                                             that grea t ly simplifies the steps required to adapt
PASCAL
                                                             software to local markets once the parent product
Theo de Klerk, 199 1 , hardbound , 748 pages,
                                                             has been released. The book fea tures a description
Order No. EY-F592E-DP-EEB ($ 39.9';)
                                                             of Digital's international product model, a scheme
Written for the professional appl ica t ion program­         for separating the core functions of a product from
mer on the VAX/VMS opera t i ng system using the             those that require t ranslation or modification for




76                                                                   1-h/. J No. 4   l'ali i'J'JI   Digital Technical]ou.-nal
                                                                                                                  I
specific markets. Also included are guideli nes for      Qual i ty. The Epilogue, Part IV, concludes with
developers working in DECWindows, VMS, and               three appendices detai l ing Benchmarking, Build­
ULTRIX environments; special considerations              ing Networks, and Networking Capabi l i t ies.
involved in preparing a product for multibyte Asian
languages or for mu ltilanguage environments; and        THE ART OF TECHNICAL DOCUMENTATION
append ixes with information o n the systems issues      Katherine Haramundanis, 1992, softbound,
in computer architecture.                                267 pages, Order No. EY-H892E-DP-EEB ($ 28.95)

                                                         Wri tten primarily for novice and aspiring technical
USING MS-DOS KERMIT: Connecting your
                                                         writers within the computer industry, The Art
PC to the Electronic World, Second Edition
                                                         of Technical Documentation has u nique features,
Christine M. Gianone, 1991 , softbound,
                                                         including i ts advice on planning and process,
344 pages with software d isk included,
                                                         research techniques, use of graphics, audience
Order No. EY-H893E-DP-EEB ($ 34.95)
                                                         analysis, definition of qua l ity, standards, and
As in the first edition, this software package leads     careers that are valuable to experienced technical
the novice step by step through instal lation, com­      writers as well . Haramundanis views the practice
munication setup, terminal emulation, fi le transfer,    of technical writing as being d ifferent from that
and script programming, and also serves as a com­        of scientific writing, and closer to investigative
plete reference work for the experienced user.           reporting. In keeping with this premise, this book
Complete with 5Y.-inch diskette containing the           is not a style guide that deals with a l l aspects of
official M5-DOS KERMIT Version 3.1 1 program from        typography and copy edit ing, but instead presents
Columbia University, t h is revision includes a new      the distil led knowledge of the author's many years
section on local area networks, additional m a terial    experience.
on running Kermi t in windowed environments
such as M icrosoft Windows and Quarterdeck               A COMPREHENSIVE GUIDE TO Rdb/VMS
DesqView, a new appendix containing tables of            Lilian Hobbs and Kenneth England, 1991,
the escape sequences used by Kermit's text and           softbou nd, 352 pages, Order No. EY-H873E-DP-EEB
graphics terminal emulators, and expanded                ($ 34.95)
descriptions of many of Kermit's features.
                                                         The Rdb/VMS relational database system was
ENTERPRISE NETWORKING:                                   developed by D igital Equipment Corporation for
Working Together Apart                                   VAX computers using the VMS operating system.
Raymond H. Grenier and George S. Metes, 1991 ,           This system is one of a number of information
hardbound, 260 pages, Order No. EY-H878E-DP-EEB          management products that work together to
($ 29.95)                                                faci l i tate the sharing of i nformation. The Rdb/VMS
                                                         system is used, for example, in high-performance
To successfu l ly compete in the next century, com­
                                                         transaction processing systems. This book is based
panies must recognize and adapt to exponential
                                                         on Rdb!VMS Version 4.0, which Digital made avail­
changes, including the d ispersion of markets and
                                                         able to customers at the end of 1990, and thus
resources and acceleration in market demands.
                                                         includes the latest functiona l ity.
ENTERPRISE NETWORKJNG: Working Together
Apart, describes how management can support
                                                         DIGITAL GUIDE TO DEVELOPING
this d istributed electronic information environ­
                                                         INTERNATIONAL USER INFORMATION
ment and move through planned transitions to
                                                         Scott Jones, Cynthia Kennel ly, Claud i a Mueller,
a new organization, confident they wil l prosper.
                                                         Marcia Sweezey, B i l l Thomas, and Lydia Velez, 1992,
Intended for individuals in charge of d irecting
                                                         softbound, 2 14 pages, Order No. EY-H894E-DP-EEB
transition of information-focused groups that
                                                         ($24.95)
extend across geograph ies, this book is segmented
into four parts. The Introduction, Part I, defines       Designed for the busy profession a l , this book
the assumptions and reali ties. Part II focuses on       presents models that extend beyond Digital and
Capabi l ity Based Environments. Part I I I d iscusses   English speaking countries in a quick read/
Simu ltaneous Distributed Work, both Goals and           reference format. Nine chapters and fou r appen­
Processes, ancl Continuous Design and Quest for          dices outl ine methods for creating written, visual,




Digital Technical jottrlla/   Vol. 3 No. 4   Fat/ 1991                                                           77
Further Readings




and verbal information for cost-e!Iective trans­           VMS FILE SYSTEM INTERNALS
lation. Primarily for information sp ecial ists,           Ki rby McCoy, 1990, softbound, 460 pages,
including writers, ed itors, i l lustrators, course        Order No. EY-F575 E-0P-EEB ($49.95)
developers, and their managers, this book wil l also
                                                           VIIIJS FILE SYSTEM INTERNA LS, based on VMS Version
help software developers and students enhance
                                                           5.2, is a book for system program mers, software
the ir background in technical communication.
                                                           special ists, system managers, applications design­
PRACTICAL KNOWLEDGE ENGINEERING:                           ers, and other VAX/VMS users who need to under­
Creating Successful Commercial Expert                      stand the i nte rfaces to and the data structures,
Systems                                                    algorithms, and basic synchronization mechanisms
Richard V Kel ly, Jr. , softbound, 212 pages,              of the v,'viS file system. Th is system is the part of
Order No. EY-F591 E-OP-EEB ($28.95)                        the VAX/VMS operating system responsible for
                                                           storing and managing files and i nformation in
This book is a concise guide to practical methods          memory and on secondary storage. The book is
for initiating, designing, building, managing,             also intended as a case study of the V.\'fS imp lemen­
and demonstrating commercial expert systems.               tation of a tile system for graduate and advanced
It is a front-l ine report of what works (and what         umkrgraduate courses in operating systems.
does not) in the construction of expert systems,
drawn from the au thor's decade of experience              DECNET PHASE V: An OSI Implementation
gained working on such projects in all major               James ,vlartin and Joe Leben, 1992, hardbound,
areas of application for American, European, and           572 pages, Orcler No. EY-H882E-DP-EEB ($49.95)
Japanese organizations. It also briefly reviews
                                                           This book provides a first in-depth look at DECnet
the knowledge representation, program ming,
                                                           Phase V and the important issues that must be
and management techniques commonly used to
                                                           resolved in the design and implementation of very
implement expert systems today, and describes the
                                                           large networks. It presents key Open Systems In ter­
intel lectual, organizational, financial, and manage­
                                                           connection (OSI) concepts and shows how DECnet
rial issues that knowledge engineers face daily in
                                                           Phase V hardware and software products imple­
performing their jobs. Among the topics covered
                                                           ment interna tional standards associa ted wi th the
are: prospecting for " legitimate" problems; fore­
                                                           OSI model.
casting costs, establ ishing project metrics and
wri ting specifica tions; preparing for system
                                                           VAX/VMS OPERATING SYSTEM CONCEPTS
"demos''; interviewing and selecting engineering
                                                           David Miller, 1991 , hardbound, 512 pages,
team members; and solving common difficulties
                                                           Order No. EY-F590E-DP·EEB ($44.95)
in design and implementation.
                                                           This book begins with an overview that centers
COMPUTER PROGRAMMING AND                                   on one visi ble aspect of an operating system,
ARCHITECTURE: The VAX, Second Edition                      terminal input and output; it proceeds into wel l ­
Henry M. Levy and Richard H. Eckhouse, Jr. , 1989,         organized chapters o n process definition, paging
hardbou nd, 444 pages, Order No. EY-6740E-DP-EEB           and memory management, securi ty, protection
($3800)                                                    and privacy; and it concludes with a chapter
                                                           on orera ting systems at D igital Equipment
This book is both a reference for compu ter profes­
                                                           Corporation. Each chapter provides an intro­
sionals and a text for students. A systems approach
                                                           duction, theoretical discussion, general ly recog­
helps the reader understand the issues crucial to
                                                           nized solutions, a lgorithms and da ta structures,
the comprehension, design, and use of modern
                                                           and questions to encourage review of the centra l
computer systems. Using the VA)( computer as an
                                                           concept presented .
example, the first hal f of the book is a text sui table
for a complete course in assembly language pro­
                                                           THE VMS USER'S GUIDE
gramming. The second half of the book describes
                                                           James F. Peters, III and Patrick ). Holmay, 1990,
higher-level systems issues in computer architec­
                                                           softbound, 304 pages, Order No. EY-6739E-DP·EEB
ture, namely, support for operating systems and
                                                           ($ 28.95)
operating systems structures, virtual memory,
paral lel processing, microprogramm ing, caches,           This up-to-elate guide for new VMS users provides
and translation buffers.                                   a sequence of steps for learning the VMS operating



78                                                                1-b/. 3 No .;
                                                                            .     Fall 1')91   Digital Tee/mica/ journal
                                                                                                                      I
system and includes hands-on experiments with            " X Protocol," " XLib," "X Tool kit Intrinsics," " Motif,"
step-by-step instructions. The book also can be          and " General X " ; all routines and data structures
used as a reference for com mands and utili ties.        are organized alphabetically within each of these
THE V1VIS USER'S GUIDE, reflecting VMS Version 5,        sections.
provides complete VMS coverage-from Jogging
in to creating command procedures; contains              FIFfH GENERATION MANAGEMENT:
a thorough discussion of files and d irectories;         Integrating Enterprises through Human
covers both the EDT and the EVE editors in detail;       Networking
and introduces programming with VA,'<TPU.                Charles M . Savage, 1990, hardbound, 267 pages,
The guide includes learning aids in each chapter,        Order No. EY-C l86E-OP-EEB ($ 28.95)
such as sum maries that contain tables of the            This book explores the challenges managers face
commands introduced in the chapter, exercises            as their organizations transit ion from the indus­
to reinforce and extend the skill s learned, and         trial era to the new era of knowledge networking.
review quizzes.                                          The author contends that new technologies like
                                                         computer integrated manufacturing (CIM) will
THE MATRIX: Computer Networks and
                                                         not be successful until organizati ons transform
Conferencing Systems Worldwide
                                                         their structures from the steep hierarchies of
john S. Quarterman , 1990, softbound, 719 pages,
                                                         second generation management to the fla ttened
Order No. EY-C l76E-0P-EEB ($49.95)
                                                         networks of the fifth generation. The book
This is the first reference book to describe in detail   contains two parts. In Book 1, " Five Days that
the extensive yet largely u npubl icized web of          Changed the Enterprise," Savage narrates a case
public and private networks and conferencing             study of senior executives co nfro nting the prob­
systems that has spread to virtually every corner        lems of a traditional organization as they work to
of the world. The first half provides extensive          transform their company into a networked
background information on the history, terminol­         organization. In Book 2, "Integrating Enterprises
ogy, standards, protocols, technologies, worldwide       through Human Networking," Savage draws on
networked com munities, and probable future              contemporary management l iterature and his own
course of networking systems throughout the              consulting experiences to present a logical case for
world . The second half describes specific confer­       his recommendations. A concluding chapter offers
encing systems and the interconnections between          ten practical considerations that organizations
them-according to geographic region worldwide.           must address to prepare for change.
Maps are included when available. Syntaxes and
gateways are provided for sending mail from one          X WINDOW SYSTEM : The Complete Guide
system to another. Additional chapters discuss a         to Xlib, PROTOCOL, XLFD, and ICCCM,
number of well-known worldwide networks,                 X Version 11, Release 4, Second Edition
includ ing the Internet and selected com mercial         Robert W Scheifler and james Gettys,
systems. Two append ices provide essential infor­        with jim Flowers, Ron Newman, and
mation on public data networks worldwide and             David Rosenthal, 1990, softbound, 851 pages,
on selected legal issues.                                Order No. EY-E755E-DP-EEB ($49.95)

                                                         By combining four MIT X Consortium standards
X AND MOTIF QUICK REFERENCE GUIDE
                                                         into one volume, this book is the most complete
Randi ). Rost, 1990, softbound , 369 pages,
                                                         and u p -to-date X Window System reference
Order No. EY-E758E-OP-EEB ($24.95)
                                                         available. In addition to the four standards, a lso
Based on the newly released X Window System              included are instructive di agrams, a detailed
Version 1 1 , Release 4 and Motif Ve rsion 1 .0, this    glossary, and a comprehensive subject-oriented
one-volu me guide combines three major reference         index. The book consists of four main parts, each
works on XLib, X Toolkit Intrinsics, and Motif           with a standard specification produced by the
programming l i braries in a compact, easy-to-access     MIT X Consortium for X Version 1 1 , Release 4:
forma t. Features include complete descriptions          Part I, "Xl i b-C Language X Interface " ; Part 1 ! ,
of approxim ately 400 XLib routines, 200 x Tool kit      "X Window System Protocol " ; Part 1 1 1 , " Inter­
Intrinsics, and 200 Motif routines. The guide is         Cl ient Commun ications Conventions Manual " ;
organized into five major reference sections-            a n d Part rv, "X Logical Font Description."


Digital Technical jourttal   Vol. 3 Nu. 4   Fall 1991                                                            79
Fu rther Readings




To receive a copy of our latest catalog or further
inJormation on these or other publ ications from
Digital Press, please write:

Digital Press
Department EEB
I Burl ington Woods Drive
Bur l i ngton, MA 01803-4539

Or, you can order by call ing DECdirect a t
800-DIGITAL (800-344-4825).

When ordering be sure to refer to Catalog
Code EEB.




80                                                   Vol. 3 Nu. 4   Fall 1991   Digital Technical journal
I S SN 0898-90 I X


Primed i n U . S . A . EY-H889E-DP/9 1 12 02 1 8 . 0 DBP/NRO Copyrighr © Digital Equipment Corpora t i o n . A l l Rights Reserved.
