The major reason for the visit initially was to run the AI/IKBS on a large VAX 11/750 configuration. As I was going to be in the USA, agreed to visit with Len Ford. As a result, DEC widened the meeting to include discussions on UNIX and AI/IKBS. As Dr Sleep, who chaired the AI/IKBS infrastructure meetings, was also in Boston, we arranged for all three of us to make the visit. We all signed non-disclosure agreements.
The benchmarks were run on an 8 Mbyte and 6 Mbyte system with a single large disc being used for file store and paging. For the interactive benchmark, scripts were run with output going to real terminals. Berkeley 4.1 was used as the operating system.
The functionality benchmarks - compiling C program etc, went through without any problems. A number of minor problems were found with the interactive benchmark. Initially it was tried with all scripts logged in as the same user sharing files. The system as configured had a maximum number of processes per user which was exceeded.
Eventually, benchmark ran in 18 mins 20 sees on both a 6 Mbyte and 8 Mbyte system. Although the editing scripts were staggered, they all caught up with each other mid-way through. The FRANZLISP interpretive program finished last.
Benchmark does not appear to test the virtual memory system too well. Given more time it would have been worthwhile redoing it but I doubt if we have the time.
Rich Ptak gave us a presentation of DEC's future plans. DEC indicated that they have had a long involvement with UNIX via Bell Labs even though they have not supported it in the past. The current commitment is to:
VNX appears to be the major thrust mainly because they want to have access to the large amount of software available under VMS. VNX will have Programmer's Work Bench, SCCS, MAKE etc. In some cases, DEC will produce similar systems also of use in the VMS environment. Main points coming out of VNX discussion:
Major points on Berkeley 4.2 release:
The AI/IKBS developments are run by Mahindra Patel who used to work for ICL on VME/B. He is very good. He left ICL with Geoff Cross and worked for A B Dick both at Chicago and Phoenix. He helped design the GEC R Machine and also did the negotiations with Roger Newey for DEC on the DEC/Manchester Dataflow involvement. Clearly, DEC must have a good idea of GEC's future plans!
Patel gave an overview of DEC's AI/IKBS activities. The main thrust is tools for two areas - the Research Community and Industrial Products. The main facilities required in both areas are access to large scale cpu power (5 MIPS) and a rich environment using high resolution graphics.
In DEC's view, the building of specific architectures such as the LISP machine is the wrong way to go. There needs to be a path through to existing algorithmic software which is denied by the LISP machine approach. Thus, AI must be integrated into the normal domain. The emphasis at DEC was, therefore, for developers to have high powered systems consisting of good workstations connected to large CPUs. At the same time, it should be possible to produce lower cost delivery vehicles for particular applications. It is not unlikely that a rule-based development system might be converted to a standard FORTRAN implementation for efficiency in the actual product.
The different characteristics of AI software were:
There were about 7 people in the advanced development group at DEC and about 12 looking at products aimed at the engineering area. This whole group was advanced development and the engineering area was used as a testing ground for new products. This group was run by Patel.
A second group under Dennis O'Connor was concerned with producing internal applications of use to DEC. This had about 80 people and was particularly looking at examples in the area of manufacturing, marketing and sales. Of the 80 people, about 10-15 had detailed AI knowledge and 5 of these could be regarded as AI experts. The budget of this group was $5M compared with 2.7M for the advanced development group.
The whole aim of the team was to produce a set of tools which gave DEC the ability to sell hardware and to allow more people to develop AI software by reducing the expertise required. The view was that the number of AI experts worldwide was 200 with 100 being in the USA and perhaps 20 in the UK.
The major development tool for expert systems being used was OPS - a joint development with CMU written in BLISS.
The equipment being used for developments was VAX 11/730 with VS100 workstation for the main developers. This was backed up by three VAX 11/780s. The view expressed was that all rule-based systems used a large amount of I/O and were therefore inappropriate for running on the workstation itself. It was necessary to have this work run on a central server. It would perhaps be feasible to move parts of a rule-based system to a workstation for processing but unclear how this could be done.
The long term aim was to have a VAX 11/780 powered system for each worker backed by about 15 x VAX 11/780 power servers.
The current team was only formed in October 1982 and would grow to about 40 people.
A second presentation followed by Mike 0' Callaghan talking about the specific expert systems developed and used by DEC. The aim was to try and increase the productivity of DEC personnel. The experts in DEC were a scarce resource which needed to be amplified.
The current group would be enhanced by an AI Technology Centre to be set up in Scotland.
The main expert systems developed and used by DEC were:
DEC were funding Edith Mumford at Manchester to work on some of these products.
There had been a split at one stage between people who extracted knowledge from experts and those who defined rules. Tending now for the same person to do both jobs. DEC had about 25 people doing that job.
The major conclusion DEC had reached for successful Expert Systems was to bound the problem area early on and not let it extend in an uncontrolled way. The number of rules in a system tended to be at least 400 and below 2,000 to be successful. However, systems as small as 100 rules were found to be useful in the diagnostic testing area.
DEC had also found that as systems got extended, the' quality of the system degraded. It was necessary at some stage to clean a system up and this would frequently reduce the size of the system by several hundred rules. Often the system became better understood in the process and, in some cases, a heuristic approach turned into an algorithmic one as a result of the better understanding.
The general impression was that OPS was quite slow and some effort had been made to organise rules so that they were as orthogonal as possible.
An example of manpower savings achieved were that DEC used to employ 28 people to deal with VAX orders. In a period, when the work had increased by 5, they had reduced the number of people to 17.
We had a demonstration of XSEL which was not that exciting and gave incredibly low response.
DEC have two A4 display products. The VS100 is already announced and I have a brochure. The second is the VS300 still to be announced. Effectively the two are 68000 based terminals with memory shared with a VAX host. The memory in the terminal is effectively part of the host address space. In order to allow it to be remote from the host, ft is connected to the host via optical fibre. The aim is obviously to get the micro VAX eventually installed in the workstation so that it will be a true workstation with the attraction of running the VAX order code. So far about 50 VS100 systems have been shipped.
The basic VS100 consists of 128 KBytes of RAM with 256 KBytes space in the VAX.. It has an optical mouse. Apparently there are problems in getting mechanical mice as SRI/XEROX have all the patents so that DEC cannot make them.
They have a good tablet made by GTCO with a 5-button cursor and controlled by a Z80.
The VT100 turned out not to have true Raster Op but just a move command as far as I could see. Alternatively, the software for removing windows was terrible. The display was A3 size with slightly lower resolution than the PERQ (same as SUN). It could move a 400 x 400 bit area in 12 millisecs. It can draw 1500 2" vectors per second.
The demo showed a number of windows being used as VT100 terminals. Redrawing windows in new positions was done by pointing where you wanted top left corner and it redrew using a most peculiar algorithm for removing old.
The VS300 is similar to the VS100 but with a grey scale or colour display. It could either be 30 Hz or 60 Hz with 16 or 256 colours. Speed was 1.37 microsecs/pixel and 19.8 millisecs for moving a 400 x 400 area. It could also output characters to the screen at a rate of 13,000 characters of a 6 x 13 font per second. There was a lot of graphics support circle, arc generation, patterned areas, line style and thickness etc.
No price was available for the VS300 while the VS100 was 9K. MIT were buying a lot and getting the VS100's to work with a VAX running UNIX. Contact at MIT is Ed Dalkovitz.
Personally, I thought the VS100 was awful. It would also be overpriced once you added the micro VAX and disc.
Sterling Forest is responsible for IBM's internal PROFS strategy. I met Gene Mazerelli who is in charge of the operation.
Sterling Forest is responsible for the production of a variety of administrative products used in IBM. Specific items include software for internal audit control of fixtures, fittings, control of manufacturing plants. It runs a corporate APL service and is also responsible for software distribution in the USA. As a result it has a whole range of equipment with some approaching obsolescence (370/168 etc). It also educates Field Engineers.
Being a large administrative unit within IBM, it was an early PROFS user and eventually, because of its expertise, was put in charge of strategic planning.
I got the standard PROFS history (there was another UK customer from CITIBANK also visiting). The current PROFS system is supported by the Dallas Office Automation Division.
The largest internal IBM user is Poughkeepsie with 1700 people while Sterling Forest is about the third largest with 400 people. A Marketing Office in Boston has 600 users. Of the 400 at Sterling Forest, 40 work at home on a special 'Work at Home' project. They access PROFS from home using software in a PC which emulates a 3101. Consequently, full screen support is available at home via a normal modern connection. When. you dial-up, the computer asks for your account and password. It then rings you back to confirm that you are at an approved location also.
The 9 people in charge of Strategy at Sterling Forest were responsible for the binary only policy as all the IBM systems were getting heavily modified locally. User exits are planned to ease the pain of no source.
Dave Robertson at White Plains is responsible for coordinating customer requirements for changes to PROFS.
The major changes to PROFS mentioned for the future were:
Other points coming up in discussions at Sterling Forest:
I also had a demonstration of document passing from Displaywriter to PROFS. All went very smoothly including inputting a NOTE this way. The visit was useful without being really valuable. The dates on when new functions would appear was very vague. The emphasis was on internal IBM support and local use with little involvement in customer problems.
The visit was completely messed up by a Tornado which hit Pittsburgh airport about an hour before I was to leave Philadelphia. As a result, I had to stay overnight in Philadelphia. Pittsburgh Airport did not open again until mid-morning. I finally got to Three Rivers around 2.00 pm.
ICL had given them a copy of the viewgraphs that I used for a comparison of systems for ICL marketing and we spent some time discussing the competition. Three Rivers were surprised that the disc speeds had not improved with PERQ Mark 2 on our benchmarks. They would like a copy to run to find out what the problem is.
Most of the time was spent with Aaron Coleman and Brian Rosen but I also saw Brad Myers (he is leaving in August to do a PhD at Toronto) and a few others. Aaron looks a typical business man and has as much faith in. Brian's timescales as I do.
The main emphasis at Three Rivers at the moment seems to be software rather than hardware. They seem to be heavily involved in mounting UNIX on top of ACCENT. They have 8 people using ACCENT which they: feel is stable. Brian had no figures for message passing times although it was clear that they had read our papers. They would like our code particularly utilities, but ICL seem reluctant to provide them. They appear to have several people at CMU helping on this. The impression given was that CMU had a larger effort on ACCENT now run by Al Spectar.
The aim was to produce a POS-1eve1 performance UNIX system by September 1983. The timescales sounded like Rosen arm waving! Having said that, they had managed to do remote file access already and IPC. A number of the major POS utilities were running under ACCENT.
By September they hoped to have the major UNIX utilities running, the SHELL and EMACS.
ACCENT had been shipped by 3RCC to Westinghouse, Siemens and FONNTEC (?).
3RCC had little interest in PNX. They had shipped about 5-10 systems. As it did not work on PERQ Mark 2, and most users at the moment were happy with POS, it did not get a great deal of push from 3RCC.
Specific answers on the software side were:
Specific answers to hardware questions were:
I did get some information on future plans but it was regarded as Commercially in Confidence - they had not even told CMU. As it was almost completely Rosen arm-waving, I am unclear as to its value.
Other points were:
INA are a large coast-coast insurance company based in Philadelphia for its central office. My visit was due to a Bell Northern salesman telling me that they were using PROFS connected to SL-1 Displayphones.
The computer centre is outside Philadelphia, about 30 minutes away, in New Jersey. It has 1,000 employees and wall-wall 3081s. It is an enormous machine room. They have MSS, 3880s etc. Interestingly they double stack cluster controllers using tables specially designed so that one goes underneath and the other on top.
They have a large SNA network with small front end systems and DISOS for document formatting.
Their decision to use PROFS was based totally on its message/note facility and the fact that so many phone calls just do not get through the time difference does not help. Incidentally, they have a telephone note facility also - not the IBM offering but a separate one.
Bob Konza, who. did the PROFS developments, was my contact. They implemented the system very quickly. They had only received the system two weeks before they offered a service. They started the Pilot with about 40 middle-senior management but found it impossible to restrict to that number due to pressure from the user population. Since last August, it has grown to 650 registered users.
Many of their executives access it from home so that they have produced a complete set of EXECs for making it user friendly in this mode. I had a demonstration and found it quite good. They will send me a copy of their manual and listings of all the EXECs.
As most of the users are management, the secretarial staff using it is small apart from secretaries acting on behalf of their bosses. There has been a mixture between those managers who use it themselves and those which are content to let the secretary do the typing.
The support of PROFS is done by their normal user support group with an on-line HELP and HOT LINE. They estimate less than 1 MY/year needed to support the 650 users.
They do not use the calendar facility because they have an existing system produced using CMS EXECs and PL/1 which is better. They are hoping to get calendar information into the PROFS users' in-basket. This is the same problem we have with network mail and they are pressing Dallas to build-out the necessary exits for them to be able to do this.
The calendar system has the multiple update problem under CMS and this is handled by a package called FOCUS which they purchased from IBI (Information Builders Incorporated).
They had experienced the same problem as us with messages destroying the screen when it would be preferable to defer. They have masked out messages where they can but PROFS is inconsistent in how it handles messages. They have been pressing IBM to sort it out.
For spread sheet work, they tried ADRS and abandoned it in favour of AUTO-TAB by CAPEX which allows hierarchies of spread sheets.
The cost of the system was estimated at about 1% of a 3081K. If DCF was used, it would be a great deal higher.
They have about 20-30 SL/I Displayphones which are all connected as ASCII terminals at the moment. By using the EXECs already written and using the functionality of the displayphone, it looks very much like an IBM screen a la Cifer emulation. The demonstration was quite impressive.
A displayphone is on order which also has a direct coaxial connection providing full screen facilities. This has not yet been delivered.
With the ASCII displayphone the user can get at a variety of facilities of which PROFS is just one. Effectively, the PROFS menu hierarchy has been implemented in their EXECs so that the four most frequently used PROFS responses at any level are loaded as soft PF keys while the others are invoked by typing an abbreviation of the command.
A second supplier of displayphones which work with the SL/ I is the CANSET system produced by a subsidiary of TYMESHARE.
The visit was less useful than it might have been in that many PROFS functions were not used. On the other hand, if they send the complete set of CMS EXECs, it could be useful for modelling our ASCII terminal support. The displayphone still seems worth pursuing.
To give some idea of the size of INA, they were negotiating with IBM for the purchase of 1,000 PCs!
SIGGRAPH was in Detroit this year - probably the worst city in the USA. You felt in danger each time you walked on the streets. There had been a substantial riot two weeks earlier.
I went to a Graphics in Office Automation workshop which took place on the first day. It was rather variable but a number of comments on products etc were relevant.
The general view on who would be major forces in complete office systems over the next few years was:
Ones that might succeed but less likely:
Ones that will not succeed:
Random comments of interest that came up were:
Hardware comments were:
There was a session on slide/foil making which made the following points:
Other minor points during the day:
The Conference was a bit lower in numbers finally than last year with only 14,454 attending some part of it and 3,470 attending the Technical Programme. SIGGRAPH made only £150,000 profit approximately compared with the 2M at Boston. It is partly due to the location but also the fact that NCGA took place in Chicago about 3 weeks earlier.
The Conference Proceedings were available to all attendees and both Julian and I have a copy if anybody wants to look through them. The papers had a very short time for presentation so there tends to be more information in the proceedings than in the presentations.
Of the Panel Sessions the most interesting were a set from Japanese companies indicating what products they had, where they were going, and their views of USA companies' efforts in Japan. The latter was quite amusing as they were quite blunt about USA representatives being under prepared and not too knowledgeable about the products and the products being overpriced.
Many of the Japanese companies were using the devices that they built and in many cases this was the original motivation for getting into the graphics market. They have products in almost every area from DEC terminal look-alikes right through to high powered coloured systems with hardware -rotation. The general impression was that more and more Japanese companies would get into this market with very competitive products. This year even, there were quite a few Japanese manufacturers at SIGGRAPH.
The audio-visuals as always were very spectacular. Speakers were displayed as they talked on a large screen. Many used two slide projectors. Video tapes were used to demonstrate techniques in papers. At the film show, laser projectors were used for high quality visual effects. It seems like half of the downstairs floor area is filled with audio-visual equipment.
There were about 200 exhibitors at the conference and several announced new products. If anything, A4 single user workstations were less in evidence than last year with only about 6 companies present. There was a great deal of emphasis on high quality colour systems - 1000 x 1000 resolution is standard with many hardware extensions to perform specific functions and at relatively low cost.
Rather than give a complete catalogue, I will concentrate on devices which caught my eye:
One of the PERQs had an attached Metheus 400 colour system which was quite impressive. The Metheus can run at either 30 Hz interlaced or 60 Hz non-interlaced depending on resolution. required (1000 x 1000 at 30 Hz and 736 x 552 at 60 Hz).
It has the ability to do pan and zoom in the hardware and also has a fast rectangular area fill operation capable of 16 million pixels/second. The cost of the display is about $9995 from Metheus direct.
Also on the stand, INTRAN were demonstrating their form design system for the Xerox laser printer. It has the ability to input fonts as a complete set and then cut them up into the individual characters.
I spent two days with Dr Manning at IBM. Geoff was spending the whole week with IBM so the first day was spent at the obligatory White Plains Briefing Centre getting an overview of IBM.
Cliff and I had a similar briefing two years ago and it had not changed a great deal. Some general facts:
We spent about 5 hours there and it was not very valuable to me. It was more use to Geoff who had not experienced the briefing before.
The second day was spent at Poughkeepsie discussing scientific systems and PROFS. The original plan had been to do PROFS in the morning but it was switched around to suit the IBM people. Then we found my plane had been rescheduled to a later time and, consequently, I had to leave earlier. The result was that less time was spent with the PROFS person.
As usual, IBM gave little away on their array processor product to be available in about two years. The impression was:
On the applications side, they had been visiting sites around the world to get a view of requirements. We agreed to get a copy of the Daresbury CRAY document which analyses users' experiences to them.
We may delay this. To get more information, we need to sign a 112 non-disclosure agreement and using the Daresbury information as a bargaining point, it may help to get the 112 signed. The IBM lawyers need to see that IBM are getting something out of it.
The PROFS part of the day was spent with Bob Rogan who is responsible for PROFS at Poughkeepsie. He developed the original code and clearly knew it backwards. He was happy to get enquiries from UK if we needed help.
They have a group of 10 people supporting PROFS. Currently have 2500 users in Poughkeepsie with 1100 in DSD. Their mission is to have all secretaries and managers on PROFS by the end of 1983 and 75% of all professionals by end of 1984. To achieve this, the group would grow to 17. By the end of this exercise, they would have 4100 connected.
They used PROFS Administrators to be the major interface to the user population. They were typically ex-secretaries who eventually gravitated to administrative assistants or even system analysts.
The site ran a number of separate PROFS systems with interconnections. This was to allow users to access PROFS from their normal processor rather than putting all PROFS users on a single processor.
Most of the new facilities mentioned at Sterling Forest originated at Poughkeepsie. They were currently moving to the standard product and were worried that they could not tailor it to their requirements once source code was not available.
Dallas had about 70 people working on the product. It was clear that Poughkeepsie did not really feel that they understood it.
Specific points that came up:
The impression is that Poughkeepsie has a wide knowledge of PROFS and if we have detailed internal queries it is the place we should go to for information. For example, they understood the PROFS title problem for Notes originating from CMS - no PROFS trailer on CMS notes. They thought that it was possible to send mail out from PROFS without touching PROFS code. You effectively send it to a specified address and pass on from there.
It is probably worth somebody visiting Dallas, AMOCO and Poughkeepsie in the future.
Note: The set of brochures referred to in the trip report can be borrowed if anybody wants, so long as they are returned to me by 26 August in time for Eurographics.