I attended courses in Advanced Image Synthesis and State of the Art in Image Synthesis, which I found extremely interesting, and mostly very well presented. Dick Lundin of NYIT would not go into details as he did not want Lucasfilm people to steal their ideas. John Robinson from Evans & Sutherland was not a good speaker, but the material (CT-5 flight simulator) was good. All the other speakers were excellent.
Tom de Fanti said that Siggraph is in very good financial shape. Siggraph is forming links with Eurographics, and the Canadian National Computer Graphics Association. ACM's rules need to be changed, currently any such relationship has to give ACM more or less total ownership, which de Fanti said was kind of insulting. Siggraph are working on that.
Carlo Vandoni spoke on Eurographics. Unfortunately he has a heavy accent, which Americans do not seem to be able to cope with, so there was much muttering, people leaving etc.
The papers presented at the conference were all interesting, and mostly well presented. I will mention a few points.
The paper on filtering high quality text seemed to be a mathematicians attempt to correct for the perverse world of physics. Owing to the use of non-linear optimisation techniques, there are not enough CPU cycles in the world to display a page of text properly.
The impact of VLSI on frame buffers promises to be interesting. People are looking towards rectangular storage schemes rather than scan-line, with RasterOp-like functions in the chip.
Ron Baecker opened the panel discussion on interaction with an attempt to be controversial: we should not be designing good user interfaces, we should be designing worlds that are easy to operate in (approximate quote). He did not get any disagreement. There were some good films, showing application of principles such as: hand movements should be directly coupled to corresponding image movements; and hand movements between devices should be minimised. Implications are things like: rate control is not as good as direct; the direction of linear movement should not change when the object being displayed is rotated.
The paper on reflectance models gave an excellent explanation why computer-drawn objects look as if they are made of coloured plastic, and how to get it right.
Nelson Max's paper explained how the use of analytic description of surfaces enabled him to vectorise the ray-tracing algorithms, and compute his images in a relatively short time on a Cray.
Tanner's paper on using a bitpad for a variety of logical input devices was useful.
Many good films were shown, some of them several times during the week. The session on Art and Entertainment contained a lot of rubbish more like home video than computer graphics. Good films were:
There was a display board at the conference for people with common interests to get together. There was one name up for graphics on Unix, but I missed the proposed meeting. Someone wanted to do hidden line removal on a Turing machine! In addition, there was a showing of sequence of images by David Em using Jim Blinn's software. He gave a good talk on how an artist works in the Course on Image Synthesis, but I did not like his taste in textures and patterns.
The Three Rivers booth was always busy when I looked. Their new demonstration is a great improvement. 3R had a Canon laser printer on show. The only competitive machines I could find were the Chromatics 7900 with a 1K x 768 colour display and a Unix look-alike, and the Superset machine, which is a bit-slice processor running Fortran and Disspla using standard graphics terminals. Text done by drawing the outline with vectors, and filling the enclosed area. It has a 48 bit wordsize, 300 KByte main memory, 30 MByte disk, several RS232 ports. Cost $27K. The Chromatics machine had a conventional dumb-terminal user interface, good colour, hardware vector generator, and zoom & pan with 3D joy stick.
There were a lot of good BW and colour displays, both vector and raster. Most of the pictures shown were done with vectors, or else digitised images. It seems that raster technology has not reached the programmers yet. Genisco had a genuine 3D display, using an oscillating mirror (yellow monochrome, slow picture update). Tektronix were showing their 41xx series - redraw speed on the 4114 is impressive. Ikonas looks like an interesting machine - high performance graphics and image processing. DEC have a 2 bit greyscale option on their VT125.
I met John Strait (3R) who was attending the conference, and of course some of the 3R people running the booth (Frank Sowinski, director of field service, Miles Barel, software engineer moving to marketing).
I met William Giarla who works for Tony Hearne at Rand. They do military work, applying Lisp and AI techniques to models of strategy and are interested in graphical representation. They ordered 10 Perqs, but cancelled when they could not get delivery. They have ordered some D0 (Dolphins?), and would be interested in Unix on the Perq. I gave him RWW's card. He said there might be someone interested in coming to work for SERC for a while.
I met Jonathan Ingram, who works for someone in Berkhampstead, and is doing a part-time degree at Brunel. He says he knows FRAH, and is trying to sell his hidden surface removal software.
William Newman introduced me to Diana Merry from Xerox PARC learning research group. Conversation revealed that there is some confusion over Smalltalk between Xerox, Alan Kay and ISI. General opinion is that Smalltalk on the Dorado is quite fast enough, and should be adequate on the Dolphin and Perq.
I met Jim Gay, Brian Rosen, Paul Newbury (now Treasurer), and (briefly) Ed Fredkin. In the software group I spoke to Don Scelza (manager), John Strait, Miles Barel, and Diana Forgy.
Brian Rosen, John Strait, someone from their hardware group, and (later) Don Scelza are forming an advanced development group to work on future products. Any suggestions welcomed. This group is expected to grow to 12 people, 8 working on architecture, 4 on design tools. Most will have a software background rather than hardware. Don Scelza wants to move as soon as possible, but has to find a replacement software manager first! Miles Barel is moving to marketing.
3R now has 100 employees. I was given a tour of the building. The PCB assembly rooms were crowded and busy. 3R have a wave-soldering machine, and a Fairchild board tester which reduces their testing time from days to minutes. The test program for the 10 board is not done yet. Current production is 30 machines/month. Newbury says that can rise to 50 in the current building. 3R are planning to move to a 100K sq ft building.
Three Rivers were running an introductory course during that week. Also they had 4 visitors from Seiko, who may become a big customer. 3R have a Japanese distributor, and 2 potential customers in Japan. Seiko are using Perqs controlling colour displays, and are also doing something with the Japanese character set. Other customers:
Brian Rosen is due in UK in October. Both John Strait and Don Scelza would like to visit SERC.
Brian Rosen's wife has just produced a baby girl. Diane Marks (Jim Gay's secretary) is very pregnant. During several pleasant evenings I met: Don Scelza and his wife Sheryl, who is a chemical engineer for a company in Pittsburgh. Don worked for Prime near Boston for 8 months some time ago. John Strait and girlfriend Peggy, who works as an operator at CMU Computer Science Dept but is really an artist. Jeff Howell and wife Carol. 3R's first ever employee. Jeff is a hardware man, claims to be John Harrington, Accounting Manager.
Everyone was slightly bemused at the scale of SERC potential order, government getting involved etc.
3R had a laser printer (Canon) running at Siggraph. They have done their own interface for it, and it should be available by Dec 81. The printer is driven from a bitmap, white space can be compressed. The controller goes on the IO option board.
Two ethernet stations talked and listened to each other while I was there.
A 20 MByte streamer tape is under development. Data can currently be written and read back. It currently runs at 30 ips, and may do 90 ips in the future. The streamer tape replaces the floppy in the existing packaging.
3R are working as fast as possible on the 1 MByte memory board (design done, PCB layout in progress) and the 16K WCS (25% of logic design done). Also more functions on CPU board to help microcoding other languages, eg indexing into register file. CMU want it urgently for the SPICE LISP development. They are developing a controller for a storage module disk, for file servers. (CDC controller I think).
The speech synthesizer hardware currently works. It is impossible to get intelligible speech out solely because of the low throughput in the Z80. The same problem will probably affect our Cambridge ring interface. NB It only needs 11 chips to interface to the Perq bus.
Intel 8087 floating point chip and Paging hardware, on IO board, or CPU option board.
Voice input. The LPC coding algorithm gives acceptable (phone quality) speech at 2400 baud, so a Perq network could offer a telephone service. (Raj Reddy at CMU).
Video camera input. Video disk/optical disk. Colour display (2 MByte system, additional monitor, frame buffer in main memory, colour Rasterop and addressing to be specified). This is rather more definite, but no realistic time estimates were given. Working on an inertial tracker mouse which potentially has 6 degrees of freedom. IO code for Z80 in ram, loaded from Perq. Reduced instruction set (don't need so many Q codes - recoding will optimise).
D.S has segment swapping, exceptions, BSI/ISO Pascal with 3R extensions, and a symbolic traceback and crude debugger. D.6 which is due out in November, will have long integers and floating point. The microcode for these already exists, compiler work is needed.
Lucasfilm have a 90% working C compiler based on Mark Williams compiler. Beware: the Qcodes will be reassigned when SALT (q.v.) is released. Lucas were going to mount Unix, but have abandoned it.
See CMU report for SPICE, Ada, Lisp.
3R were going to contract Advanced Computer Techniques to implement a Fortran compiler, but turned it down when ACT increased their price to $100K.
3R are considering a proposal from Human Computing Resources Inc. Toronto (Ron Baecker) to implement Bell Labs Unix V7 on the perq, price $100K.
3R are very keen to have a fully supported Unix available, and want a third party supplier to do this.
3R expect to release SALT aka SPICE 81 aka Version E in Mar/Apr 82. (See CMU report). This will offer multitasking, IPC etc. 3R are devoting a lot of their manpower to aid CMU in achieving this, but will not make a commitment to release dates until they see something run. CMU intend to put all their software in the public domain.
SALT will have a better stream/window package, and the CANVAS graphics system. The SCRIBE document editor from CMU will be available. 3R currently use PROSE, written by John Strait at the University of Minnesota, who are now claiming ownership, so 3R cannot release it.
A software documentation and control package called VIRGIL is running in-house, and may be out with D6.
ARPA IP/TCP protocols will be implemented for the ethernet.
CMU will develop an Ada compiler when SALT is running. Currently a subset (Q code semantics) compiler is running on the VAX at CMU.
Incremental backup utilities will be available for the streamer tape. Spice will offer archiving facilities. Long term developments will probably be initiated by the Spice project.
I first spoke to Phil Hayes, an Englishman working on the research faculty at CMU (via Oxford, Edinburgh, and Rochester). Phil's research is on user interfaces, now known as Cousin (Cooperative User Interface). He uses an interactive mail system as the domain of experiment(see document). The original system involved multiple processes on a VAX with an Alto front end, and was far too slow. It is now being redesigned for the perq, making more use of graphics, and further separating the interface from the tool. The first implementation used a mainly textual interface, with a helpful and tolerant command analyser, and used the Alto mouse occasionally to resolve ambiguities by pointing at the desired element.
Peter Hibbard is the project director. Spice originated as a proposal in 1979 for development of personal scientific computing for research work, including CMU computer science. See document. CMU were prepared to invest 100 man-years of effort to re-implement the current state-of-the-art in programming environments. Spice is intended to support large programs and projects - eg the COUSIN knowledge-based user interface.
Main workers on Spice are Sam Harbison, Rick Rashid (Accent, q.v.), Gene Ball (graphics, programming environment, user interface). Scott Fahlman and Guy Steele are working on Lisp (q.v.). Spice is intended to be portable - the Perq is being used because it is the closest of the available systems to the original requirements. Other hardware may be used in future.
SALT (aka SPICE 81) is seen as a vehicle for developing Spice. Accent (the kernel of SALT) provides a message based inter-process communication primitive.
Accent messages are structured, and contain explicit identification of the type of fields. (This makes open systems interconnection simpler and more reliable). An implementation of Accent is currently running under Unix.
SALT features are:
It is CMU's intention to put SPICE software into the public domain. They expect to make SALT available to 3RCC in September 81.
SALT will have the necessary support for ZOG (USS Carl Vincent will contract 3R for support). SPICE Lisp etc. Paging microcode will be available for machines with 4K WCS.
Guy Steele is working on Lisp to run on Spice. Scott Fahlman is project manager. Spice Lisp is a descendent of MACLisp. and will be the common Lisp which is to be implemented on a range of machines. It is approximately a subset (70% functionality) of Lisp-machine Lisp. The speed is estimated at half to two-thirds of a KA10. Much of the programming environment features offered by. eg Interlisp. will be obtained via the Spice environment.
The compiler is currently being debugged. A simulator for the order code runs on a VAX. (Spice project have a perq microcode simulator written in Lisp - if in doubt, recurse!). Most of the Lisp microcode has been written. Pieces have been tested on the Perq 1 MByte + 16K WCS is needed. The system uses a stack for function arguments where possible. Upward funargs are implemented on the heap. Lexical and fluid bindings are equally supported.
They expect to have something to release by the end of the Spring term 82, but it depends on the common Lisp effort.
Prof Raj Reddy runs the distributed sensor net project and some robotics projects. The distributed sensor net (DSN) involves a number of computers cooperating (using AI techniques) on lOGating objects given input from a large number (100 to 10.000) sensors. The application is locating submarines with sonar etc. The test setup is a toy train set on the floor of a quiet room with several microphones scattered around using time difference of arrival of signals. The DSN project will use Spice and DPL-81, a distributed processing language featuring static definition of communication paths, fault tolerance, and dynamic reassignment of processors. Work is being done on distributed debugging using a pyramid structure to give different levels of abstraction.
In the tour of the robotics labs we saw: a Perq detecting errors in pc boards using a vision system; a one-legged robot with balance (not operating); a six-legged walking machine; a project on multi machine cooperation (one machine picks out an object, another positions it accurately etc); and a direct drive arm (motors in joints - no gearing system).
The Spice project has an industrial affiliates plan - $100K per year to participate. VAX 11/750 less powerful than Perq. Dorado not available outside Xerox. Dolphin probably less powerful than Perq, Star about equal to Perq. William Wulf is leaving to form a "Tartan" company. Bob Sproull rumoured to be leaving to form a graphics house - unconfirmed. Gene Ball has a voice recognition device but so far has only played with it. Spice will eventually include voice 10. CMU have about 30 Perqs, 20 Altos, a DEC 2060 and a KL10, 2 KA10's, and more VAX's than a man can count, all in the Computer Science Department. It was an early policy to have high computer power per person. CMU have 3 MHz ethernet on Perqs. They are building a 1 microsec clock for Perqs. The DSN project ran some algorithms on Perqs and VAX. perq is about two-thirds of VAX performance. When they microcoded the inner loop, they obtained 2 x VAX speed. Dana Scott is now in a permanent post at CMU. Whitesmiths Unix runs OK on M68000.
On this visit I was accompanied by Jeff Wilson (see Apollo trip report). Computer Science is a small department (7-10 faculty). Leading faculty members are Andy van Dam (working on his book, Principles of Interactive Graphics, supposed to replace Newman & Sproull), Peter Wegner (interested in Ada, wants to do computer aided instruction, possibly on Apollo), Steve Rice (C cross compiler for Apollo from VAX, data bases), Gene Cerniak (Artificial Intelligence, Lisp), Steve Feiner (Interactive document system, presented at Siggraph 81 q.v.).
Brown Univ currently have 5 Apollos, 3 of them networked. They plan to buy 40 nodes, set up in a lecture theatre similar to a language teaching laboratory. Computer Science students will use an interactive programming system (syntax-directed editor, incremental compiler, source level debug) during their lectures. Communications allow the lecturer to monitor students work, give examples etc. Graphics will be used to show animation of algorithms, highlighting source statement being executed etc.
They plan to use Unix as the basis, initially implemented on top of Aegis. MIT's 68000 C compiler is currently running on their VAX under Unix. They expect to have 15 nodes by Sept 82, increasing to 40. They plan to have 2 students per node, possibly with 2 CPU's (1). They will build a gateway to the VAX, and to BRUNET, their campus 300 MHz network currently under development. They are considering implementing Smalltalk, but are held up by need for documentation and licence. Apple, Tektronix, HP, and DEC are reported to have Smalltalk licences.
Apollo's distributed filestore is good. There is no interprocess communication, no distributed processing, no RasterOp. Erasing a line by software is 10 times slower than the primitive which draws a line (which I believe is written in Assembler). Early benchmarks show that remote file access is 15-20% slower than local. Array access is about the same speed as a Perq (I don't know how they determined that).
They have a graphics package in C which they consider to be a 2D subset of the GSPC core. They are upgrading for 3D, and want to add Raster primitives. They are interested in GKS multiple workstations.
Brown Univ considered Western Digital's NU machine and 3RCC Perq to be unreliable (university design) and too late. Xerox Star had no programming language. They want to develop a special relationship with Apollo. They have some complaints about the Apollo software (eg no Pascal yet) but will wait for the November release before making a fuss.
The demonstration of their interactive document editor and viewing system was good. They use colour graphics via a Z80-controlled Ramtek monitor. Their picture editor is flexible and has a good user interface.
The general impression I got was that as a small department, although they are enthusiastic and expound the right sort of ideas, they will not make any great forward progress in interactive environments.
MIT have had 2 Apollos since May, but have only been working with them since July. Both have 1 MByte of memory on four boards. They have implemented an FTP in Pascal which uses one of the RS232 ports to communicate with their DEC 2060. MIT have lost their latest source, but Harvard and Yale have copies! They consider the Apollo software to be rudimentary but adequate. They have no particular interest in Unix. They like the transparent file store, and intend to use the multi-bus interface (see CalTech report) to build a gateway to Chaosnet (broadcast, interfaces direct to processor, no buffering, throughput 1.5 Mbits in 5 sec).
MIT really want NU machines - expect to get good discount. They expect Western Digital to have re-engineered the NU by abut Feb 82, probably using the 68000. The original target price was $20-30K. In the meantime they expect to buy 20 or 30 Apollos.
Apollo performance appears to be approx 1/6 of an empty DEC 2060, or about equal to a loaded 2060. The Apollo screen editor is claimed to be slower than EMACS on the 2060. In essence, Apollo was chosen because of availability. Apollo is twice as fast as the original NU, but has 4 times the clock speed. There are rumours of a single user VAX from DEC, but target is 1985.
Chris Reeve is implementing MDL on the Apollo. MDL is a Lisp-like language, with a machine-independent compiler. MIT intend to mount it on the VAX also as a portability exercise. I saw some trivial MDL programs run on the Apollo. No major applications have been ported yet. MIT wanted an assembler and debugger, but Apollo did not want to release theirs. Apollo compilers and system software make assumptions about the program's environment, eg register assignments, control stack structure. MDL has about 6K Bytes of assembler code, for interfacing to Aegis, and handling procedure entry and exit.
DL is aimed at 2 main projects: Office Automation (see later) and Programming Technology Group's planning system, which is a knowledge-based system which aids users in collaborating on project planning. A knowledge representation language is used to provide help appropriate to the context, as well as the usual network mail functions.
Richard Ilson of the Office Automation Group (2 staff, 3-4 students) is implementing the Etude document editor (see document list) in MDL for the Apollo. Etude was originally written CLU on the 2060, can use standard terminals, but needs a bit map display to handle multiple fonts properly. Etude offers a good user interface, English-like commands, undo facility, interactive editing and formatting. They want to use a language which is available on both the 2060 and whatever workstation they buy. The NU machine ran CLU, but they are not available so they are re-implementing in MDL for the Apollo. They are extracting the user interface to make it part of the environment, along with the window manager, where it will be called ECOLE.
They plan to implement a page layout system using similar techniques; a personal/office database system with query language, a Visicalc-like table editor, and business-style graphics.
Other projects are Office Analysis Methodology (basically common sense). Office Specification Language, and Multi-person information work (eg Calendar system described in a recent publication).
This visit, along with Brown University and MIT, was organised by Jeff Wilson, Product Marketing Manager for Apollo. Jeff accompanied me on my visits to Brown and MIT, for his own interest in seeing them (he said). Jeff is in his 30's, ex-DEC corporate product planning, where he was not allowed to get his hands dirty. He is building a Western Digital Pascal Microengine + Z80 video terminal in his basement, which is not working yet.
I talked to Mike Greata, Vice President Engineering, and Andy Markowitz, who is Apollo's graphics software expert. I was shown around the development labs, where I saw a number of Apollos in use for software development - support tools are being ported from their Prime to the Apollos by stages. There was a black and white monitor in a test rig with a few characters (white on black) on the screen, but I did not see it being driven by an Apollo. They say they are working on changing the packaging to fit it in.
There was a touch pad (about 3" square) also not functioning but connected to some electronics. They expect resolution to be about 2 pixels, possibly 1. To signal a hit you will have to press a button - supposedly with the pad mounted in the existing keyboard the "mark" button and some others can be pressed with the thumb while pointing with the index finger. I suspect that two hands will often be needed. The pad works by measuring the resistance between opposite edges of the substrate and the conductive layer which is pressed onto it by the finger. This is done repeatedly in x and y directions alternately, and some averaging is needed.
I saw a 1 Mbyte memory board being tested on a dismounted Apollo card cage. That is expected to be available soon.
I enquired about the effort involved in replacing their network with a Cambridge Ring. After some talk about gateways, we got down to the real issues. I got a definite impression that the Apollo token ring is closely integrated into the system hardware and software, that it would be a major effort to change, and that they really did not want us to do it. Poduska (see later) said that Apollo were really selling the network, not the machines, as they want to be able to build new architectures for the nodes without making the DOMAIN system obsolete.
They intend to build a Colour version with 1K x 1K resolution (1K x 768 viewable, interlaced, colour or monochrome monitor), 4 to 8 bits per pixel. Each pixel plane can be mapped into the processor's address space - one at a time. RasterOp hardware will operate plane-by-plane.
A node with no disk (for cost reduction) will be available in Nov 81. They want to achieve less than 20% overhead on paging, but will settle for 2:1 degradation initially. Currently they expect 6% overhead for paging code (ie read-only) across the network.
Processor upgrades: 10 MHz 68000 are available now, 12 MHz on the way. National 16032 under development, specs are available, high speed 32 bit VAX-like processor. Intel 432 is not hopeful: currently it is slow (~ LSI-11), floating point will be 8087 but slower. Also Fortran will not map well onto 432 due to things like 64 K Byte limit on object size. Apollo intend to build a CPU enhancement board, with Intel 8087 floating point chip, and cache memory. With this board, Mike Greata estimates the Apollo will be between Prime 250 and 550 in performance.
First magtape drives should be available in Sept 81. General availability with save/restore/interchange utilities in Nov 81.
3-600 lpm Printronix printer should be out in Sept 81.
66 MByte disk coming "soon". 154 MByte 14" Winchester will be available next year. 300 MByte storage module controller is under development.
Apollo intend to release software on magtape in the future. Multibus interface (single master) promised by end of Sept 81.
Currently Fortran 77 and SPL are available, with software tools (Ratfor = Unix V6). SPL is essentially Pascal without floating point or 10. Pascal will be available in Sept 81. Apollo are talking to Precision Visuals and Megatek about buying a GSPC core implementation. This is CONFIDENTIAL. Colour will be available almost as soon as the hardware is available. Meanwhile crosshatching will be used to simulate colour. The assembler is not being released, but "responsible" customers will be able to get it by asking.
Pascal and F77 will conform to standards, and have extensions (eg long names, Include statements). As source level debugger will be available by April 82.
The display manager will support variable width windows, multiple fonts and graphics, and the touch pad "in the near future". Many fonts are available from Brown University (presumably from VAX Unix). Apollo are "actively interested" in interactive programming languages integrated with the shell. Also features equivalent to Unix V7 shell will be out next year. Data management package end 82. Apollo are going to put together a group to think about longer term plans. Currently they have no plans for typesetting, tools for interactive interfaces, software management. They expect researchers to show them the way. Yale are working on a computer architecture design system. Some others (unspecified - confidential) are working on PCB and gate array and logic design software' on Apollo.
After I had left Apollo that evening, Bill Poduska called me at my hotel. He was sorry he had missed me etc, and was eager to meet me before I left. We met on Thursday morning for breakfast before my flight to San Francisco. We had a general chat about network/single user system environments. He emphasised that Apollo wanted good relations with SERC. He told me that the day before (Wed 19th) Charlie Specter and Gerry Stanley had met Wilmot at Boston Airport, and discussed a marketing deal. Wilmot was interested in manufacturing. ICL plan was to get into the market by (in order) marketing, manufacturing, and then engineering. Poduska said it was a decision between 3R and Apollo. I played dumb but interested.
My initial contact was Karen Bedard (via Brian Rosen), thence Bob Gaskins (her boss - Manager, Computer Science Research) and Pat Milligan. BNR have 2 Perqs, which they claim were among the first orders. They were disappointed by the late delivery and lack of software.
Their research is directed towards a personal workstation as a communications station - replacing a telephone by (voice + data + video). They are interested in using graphics in conferencing, and showed interest in the multiple workstation facilities of GKS, which they had not heard of.
BNR plan to develop their own hardware for their workstations, based on Motorola 68000, using the Intel 432 for processor power if needed. Their reasons are cost (they do not need many for their own research) and availability of Unix. They have a 432 development system. ADA compiler runs on a VAX in Boulder, Colorado; code is sent to a MDS, and thence to the 432 system consisting of 4 memory boards, 2 10, 2 processors (which can run as transparent multiprocessing or as duplicate processors for reliability). The edit-run cycle takes about 30 min. The 432 is slow, but they expect it to be sufficiently fast by the time Intel get full production going (estimate 18 months).
Karen has a menu package for the Perq, including a menu editor which enables one to draw a picture, specify buttons etc. Also a program for proof-reading the device-independent output from TEX, which runs on their DEC20 (SAIL). They have some fonts for the Perq generated by Metafont (also on the DEC20) which are very poor: Metafont works badly below about 200 dots per inch.
Pat Milligan has developed an interpreter for the Smalltalk-80 virtual machine in Perq Pascal. They are waiting for the tape from Xerox to test it. Generally they seemed happy about the idea of swapping software with us, but they do not plan to do any major development themselves.
Other systems BNR know about:
Xerox are offering the Dolphin (as Xerox 1100) with InterLisp, Smalltalk in the future. Xerox salesman says if anyone orders and possibly more than 50 Stars, they can get Mesa, and Smalltalk sometime next year.
Stanford University Network (SUN) have 68000 based workstations, no virtual memory, plan to implement Smalltalk.
BNR were very impressed by a demonstration of Lilith (designed by Wirth at Zurich after a Sabbatical at Xerox PARC: 2900 series bit-slice processor, bitmap display, microcoded RasterOp, Modula 2 single language system). Good graphics, multi-window source-level debug, good performance. Liliths are being built to order by Brigham Young University, contact Richard Oram.
Someone called Birnbaum has moved from IBM Watson Lab to HP Labs. HP plan to reduce research staff to about half (to clear dead wood) and then build up to about four times current level. They are raiding Xerox PARC - already have Alan Kay and Ira Goldstein. Rumours of Bob Barton going there also.
BNR claimed that Sprocket systems had a lot of reliability problems with Perq monitors - this turned out not to be the case.
They gave me a listing and a tape of the ARPAnet mail on workstations.
We spoke to Henry Thompson (now at Edinburgh) and Martin Kay who work with Interlisp, and Jim Horning, Jim Mitchell and Jim Murray who work with Mesa. We saw Dolphins and Dorados. Dorados are claimed to be 10 times faster, and looked it when comparing animated towers-of-Hanoi programs. Apparently several bugs appeared on the Dorado owing to single key depressions being seen as two hits (the keyboard and mouse buttons are fully decoded, and the processor samples the current state). The problem arises from the system reacting too quickly and inspecting the keyboard again before the button has been released. I call that a bug, not a timing problem.
The Dorados are normally kept rack-mounted all together in an air-conditioned room. Otherwise they have to be mounted in a large box ("armoured personnel carrier" in PARC terminology) because of heat and noise. There is no sign of Dorados being available outside Xerox.
Getting the entire Interlisp environment to run efficiently on a new machine is considered to be a major undertaking. Interlisp is clearly a mature system, offering a rich environment. However, in terms of man-machine interaction, we saw nothing that has not been done in prototype on the Perq. Windows do not slide across the screen, only the outline. Scrolling, thumb bar etc are just like the Perq. Of course, multiple tasks in different windows, saving environments, loading new microcode from shared file servers all work.
PARC people obviously think highly of their Mesa environment, and consider Office Products Division wise to use it for the Star. I have heard outside opinion that it is not that good. There is a definite agreement among Mesa users that ever more rigorous type-checking etc is a good thing. Nobody had much to say about the Perq. Unfortunately we did not talk to any of the Smalltalk group.
My overall impression was of a lot of good ideas. but not much of a coherent framework for creating a complete environment.
Lucasfilm currently have 5 Perqs. Only 2 are in use on a video editing system; the other two projects planned for the Perqs are using other hardware, since they could not develop the software in time on the Perq. They want to sell 3 Perqs as soon as possible, and the other 2 after a few months. They are not at all unhappy with their Perqs: they are building their own systems based on 68000 and multi-bus. They claim they can make all they need cheaper than Perqs, and can port or develop the required software faster. They use C, obtained the 68000 compiler from MIT. They do not use LISP at all. They have Bell Labs Unix V7 running on a 68000.
Tom Duff has developed a C compiler for the perq, based on the Mark Williams compiler (licence needed from Mark Williams). This runs under POS, but is not fully tested, does not have long integers or floating point, does not properly handle copying structures on procedure calls, and does not completely support initialised variables. The compiler may well not be compatible with the F77 front end. They intend to give their compiler to 3R in exchange for a Perq source licence. When Sprocket abandoned Unix development on the Perq, 3R wanted up-front money from them to pay HCR to do it. Sprocket recommend that 3R should get a binary licence from Mark Williams so that they could offer Unix at low cost. We should keep in touch with Sprocket on their compiler work, but release will be via 3R.
The video editing project uses a Perq to control 2 or 3 video recorders with two monitors. Using a good menu system (pos Pascal), they can create, edit, and store a sequence of video edits. These are specified by, eg, play tape 1, pause, mark this frame as start of sequence, etc. with monitors showing the picture. The resulting effect can then be reviewed by replaying the edits. I have attached a print of the Perq screen during editing. The menu package is very slick; it was written by an ex PARC man. Since they will not use perqs long term, they will probably let us have their software. I left a floppy for them to put it on for us.
They are using the perq keyboard and screen on 100 foot cables to get away from the fan/disk noise. They only had to change one resistor. Also they have used a GPIB/Unibus interface to communicate with a VAX.
Other machines: They have looked at the Chromatics, which offers colour and Unix look-alike, but no ethernet, to which they are committed. Apollo costs as much as a Perq and is not as good. MIT NU-machine has Unix but a special bus. Perq is short of 10 Ports and has no floating point currently. Most 68000 workstations are unsuitable: they really want a general purpose computer.
CalTech have a small (2-3 people) experimental project to develop a low-cost colour graphics frame buffer for VLSI design. They will use Apollos to control the frame buffer, and have 2 for this purpose. They do not plan to buy more for a year or so. The frame buffer system will consist of one to three boards each containing 4 pixel planes (640 x 480) and a graphics processor (NEC 7220) for each, and a colour table board with up to 4096 x 24 entries. These will use the multi-bus for communication to the host (should be available from Apollo end Sept). They would like to attach a peripheral processor to the Apollo, but the multi-bus implementation will only permit a single master.
They are preparing to make their first wirewrap boards - hope to have a working prototype by the end of the year. They are considering how to release to outside users. They expect the system to be compatible with the SUN workstation. They may licence the design to some multi-bus manufacturers, or else sell unpopulated PC boards. The graphics processor will provide functions such as line, circle, rectangle fill, circle fill, hardware zoom and pan. They are currently using Fortran, but will use the Brown University C compiler. They expect most software to be portable, but the user interface will be machine dependent.
Other groups are interested in Smalltalk, possibly on Perqs. Three Rivers have sent a salesman there.
Apollo speed without floating point is approx 1/3 of a VAX. A = B + 1.0 takes 150 microsec, multiply more like 250. Operating system response is not as good as VAX Unix with 4 users. Continuous text output to the screen runs at effectively 6600 baud. They have used the network up to about 1,000 feet, and the speed is OK. They have not made much use of multi-tasking as everything slows down. They suspect only one task can do IO at once (ie interleaved at block level). There is no cursor hardware - even the text cursor operates by saving the character being ever written. They have lots of fonts from Berkeley Unix on the VAX.
ISI are Dept of Defence funded. They work in Computer Science, Languages, Communications, Protocols, Interlisp development etc. They are a support centre for military research and operations for Strategic Air Command, Airforce, Army etc. They act as a test bed for future government procedures, especially in local area networks, and portability. They currently use 6 DEC10's, several 11's, and one VAX with Interlisp. They support approx 3,000 users. 90% do no programming: write documentation, memos etc. 10% work with major Interlisp applications. Everyone at ISI (including the receptionist) has a terminal (HP 2648) - all will get a personal workstation.
ISI want to use personal workstations because of: large address space (Interlisp), good interaction (document production, mail etc) and resource sharing and communication.
ISI major groups:
However their only required language for workstations is Interlisp!
ISI considered SUN workstations, Xerox Star, MIT NU, perq, Dolphin. They are choosing Perq, largely because of the CMU SPICE effort. They will use VAX's as central file servers, gateways etc. There is still a faction in favour of Xerox Star - mostly Interlisp users who want a DEC10 each by getting the others onto Star. I suggested they would be better off with a network of Stars and a few Lisp machines, which they agreed would fit their needs, but they want to avoid having to support common protocols on a number of different machine architectures. Also I suspect there is opposition to changing from Interlisp to MACLisp - SPICE Lisp will be close enough, although they see a good Lisp environment taking a long time to develop on the Perq.
ISI have 1 perq, and 2 more on order. They will not order more until they definitely decide to drop Xerox. They expect to buy about 100 Perqs over 4 years. Xerox Star is currently their second choice. Xerox will not let them write microcode, but may give them access to the bytecodes. Xerox have offered Mesa, and will develop Interlisp for them. Total price comes to about $32K (extra memory $5K, Interlisp 5K/machine, Mesa $5K, Op Sys $1K, basic price $16K) so it is about the same as a Perq, but has good office automation software. Xerox have offered Smalltalk in 6 months; Xerox EOS will offer Interlisp.
ISI have ordered 2 Dolphins, but consider them overpriced at $60K. They may get 1 or 2 Jericho machines from BBN. They have a Penguin (Xerox 5400 Printer) but cannot get service as the Xerox organisation is in chaos.
Note that Star is a software system which runs on the Mesa/Pilot environment, which runs on the dandelion hardware (Xerox Star as advertised).
