Jump over left menu
USA/Canada Visit, August 1977 by Bob Hopgood (and Cliff Pavelin)
The main reasons for the visit were to attend the inaugural ISO Computer Graphics Standards Meeting and IFIP77 in Toronto and to visit Prime near Boston>
The order of the visits were to be:
- August 2: University of Toronto
- August 3-5: ISO TC97 WG Meeting
- August 8-12: IFIP 77
- August 13: Prime Management
- August 15: Prime
- August 16: IMLAC (Bob Hopgood)
- August 16-18: Cliff Pavelin visited Prime customers at Michigan, Baltiimore and Rennselear
The flight to Toronto was uneventful although I (Bob Hopgood) had problems with Customs on arrival. They thought that two films I had brought in with me were classified as merchandise connected with my trade and that I should have declared them rather than waiting for them to be found when they searched my baggage. They warned me not to do it again and threatened me with confiscation, prison etc. Other than that...
2. University of Toronto
On Tuesday I visited Ron Baecker at the University of Toronto. They had had a bad fire a few months before which had burned it nearly to the ground but with the loss of not much equipment. The building itself was in a sorry state.
They have a PDP11/45 running a modified Unix system which was supporting about 6 users without too much trouble. Their main peripherals are a display from Three Rivers Computer Corp based on the Hewlett Packard tube and a Sony video system with a 256 by 256 frame buffer. Both devices have their memory looking like genuine PDP11 memory. The display company appears to be an off-shoot of Carnegie Mellon and has Stan Krist as leading hardware man. The display was not very fast. It can generate 50,000 short vectors flicker free. We were able to put up a full page of software drawn Gothic characters having 30 to 40 vectors each. They were using this with a Sumagraphics tablet. The video system was not so impressive in terms of software that they had available. I got the impression that it was a recent acquisition and not too much work had been done on it yet.
All their software was device independent working with the display, Tektronix, Versatek, lineprinter etc. The input side seemed a lot less device independent than the output. A lot of the graphics software had been added to the Unix operating system itself and seemed very efficient especially menu picking even though it was being done by software.
They had a number of projects being done either for Master's or PhD's. The most interesting for me were:
Source Graphics Debugger: worked rather like DDT but you were able to scroll up and down the program until you wished to put a breakpoint. This could be placed on any quantity including a part of an expression evaluation just by indicating that point with the tablet.
Scrolling could either be on a routine heading or source basis. Thus you scrolled until you found the right routine and then scrolled in the other mode to get at a particular line.
Page Layout or Newspapers: they are hoping to market a complete system. The data for the program consists of articles under a number of distinct headings with their size in columns. The system allows you to make up the page selecting stories in particular categories until the page is to your satisfaction. You are allowed to change the size of photographs to fit in with the text etc.
Animation System: based on GENESYS, it had all the usual facilities such as P-curves etc. The only real novel feature which I don't remember seeing before was a method of avoiding the discontinuity when inbetweening between keyframes A and B and B and C at the position B. Normally you interpolate from A to B to get the inbetweens and do the same from B to C.
In this system, you are able to merge the two by extrapolating backwards from C to B (past B) and averaging this with the interpolation between A and B. The result is a much smoother inbetween near B although the B key frame is now never drawn but may be some average depending on the way the interpolation is done.
The graphics software as a whole was nicely thought out with a number of useful dodges.
The cursor used for picking changed in shape depending on the state of the system. While moving it about, it had a diamond shape. When you hit a light button, it turned into a Buddah while it was computing the next action (Buddah = thinking). If the attempted action was currently illegal you got a thumbsdown cursor. If you lost the cursor through moving the pen off the paper, a query came up. When the action expected keyboard input, the cursor turned into a keyboard.
The idea works well. You are almost certainly looking at the cursor when the action is performed, thus it is the obvious place to return the prompt.
Menu hits were split into two classes. One class was obeyed immediately while the second class required a second operation such as a button hit to confirm it.
There was not a lot of new animation work. Ron appeared to be less interested in it now. He was still producing his Sort film showing sorting techniques and also one describing different Head seek algorithms for disc drives. Both did not seem to have changed much since last year.
A new film did show different algorithms for a fleet of taxis replying to calls to optimise different parameters such as miles travelled, wait time of passengers etc.
Their PDP11/45 had CDC 40 Mbyte drives which they had purchased OEM from another company. They had had endless trouble with the controller and service and vowed they would never do it again.
- Brief User Manual for Animation Package. Complete thesis to follow.
- GPAC User's Manual - their computer system.
- Six papers on animation by Ron Baecker produced for a variety of local conferences by Ron since 1973.
- Publicity blurb for his Company: Human Computing Resources Company.
I showed them my Hash Film and the FE Film. They were impressed by the sound track to the Hash Film - they have a synthesiser and one student doing a PhD on the psychology of sound, teaching etc.
3. ISO Meeting
A large part of the first day was spent on procedural matters. The Working Group had been formally ratified by the TC97 Committee (That is two levels up in ISO). It was evident that the USA contingent were interested in what would happen to the GSPC Core proposal when and if it was placed formally before the Committee and before that the Working group. They wanted to know the difference between members and correspondents, who were allowed to vote, what a majority vote is etc.
To some extent these issues are well defined by a document for the higher levels of ISO but not at the Working Group level where the particular group is allowed to formulate its own code of practice and have it approved by the higher body - in this case TC97.
The meeting re-affirmed that its aim was to see some level of standardisation within the next two years and that this would be first in terms of functional capabilities and secondly in terms of FORTRAN subroutine calls. There was some argument over a UK proposal to focus on 2D Level 0 graphics at this stage. In general the view was that a decision on this should be left until the third day.
3.1 Standards Activities
It was evident that the main proposals for a standard would not just be the GSPC Core but possibly also proposals from Germany and Norway. The Germans have taken the Core proposal and have defined a Core system from this which will go through the local DIN standards committees with an aim to getting internal approval within Germany.
The ANSI/SIGGRAPH and German groups are corresponding and will attempt to keep some kind of compatibility although it is clear that there are some differences of opinion. The Germans need a local standard and that is their first objective. They may submit it to ISO as a possible international standard but are not committed to it as ANSI are.
Norway are currently redefining GPGS-F so that a clear distinction can be seen between the modelling and core parts of the system.
3.2 san Jose SIGGRAPH Meeting
There was, to some extent, a divergent view of the results of this meeting from the one provided by Alan (Francis). It sounded much more of a sparkling success with a broad measure of agreement.
The original GSPC Working Party has to a large extent stopped and is being replaced by new faces. The main reason is that the people involved are tired out. They estimate 40 man years have gone into the production of the Core Proposal.
The current attitude is that the proposal is in a state where implementations can be carried out and these are already underway at CDC, Los Alamos, Tektronix, IMLAC, DISPLAA, Vector General and NASA. Little information was about the level of implementation or the commitment involved. My own view was that it was a political statement aiming to steamroller the current proposal through. The ISO meeting was a bit unhappy about the undue haste but agreed to leave any decisions to the last day after the formal presentation and comment of the Report on the second day.
The second day started badly with a long and detailed discussion on the basic output primitives on similar lines to our local discussions. There was yet another strong attempt to throw out Current Position (CP) all together and other attempts to have multiple CPs. The eventual consensus was for a single CP with a strong recommendation that something should be done to clean up the position of CP after low-level text.
GINO-F and many other systems that contained both basic graphics and modelling used Current Position in an ill-defined way. An IFIP Workshop at Seillac in France had already proposed that all primitives should be absolute with no CP.
There was a recommendation that the POLYLINE primitives should indicate a move to the first point and a draw line from there to make them better defined in terms of both position and shape. It was clear that the whole set of reasons for advocating POLYLINE as a primitive was because BROKEN attributes had not been considered by GSPC.
There was also strong disagreement with the current level structure and an attempt would be made to redefine these in a more sensible way for storage tubes. Dick Puck of GSPC would coordinate these activities.
There were a number of proposals that arose in connection with MARKERS including a suggestion that they were just another line attribute!
The day was fruitful in that the GSPC members defended and gave their reasons for particular recommendations:
- There was a great deal of discussion over input devices and the separation into Events and Samples. A number of alternatives were raised but there was more support for the current set than against it.
- There had been a possibility of a pseudo device being defined in the GSPC Core. This was strongly urged for by the Germans and it looks as though this may be added.
- Copying segments, even straight copies seemed to be outlawed because of the difficulty of doing it other than by interpretation when, for example, the segment had been fragmented in memory due to updating etc.
- The two-level naming was defended as a useful PICK-ID return for the reason that the segment name was also useful on a lightpen hit of a menu item as there was a possibility that the segment would need to be changed in some way. There was a discussion on whether segment naming was just a file handling mechanism in some sense and should be disjoint from the picking function.
- It is possible that line attributes will be made indirect so that you indicate that you want a linetype of 5, say, and another routine tells you the format for this type.
- Strong claim that the Core should not be used for outputting error messages of the Core.
- Image transformations were, it was felt, needed for efficiency on all devices. After having clipped an object, there was a need, even on a storage tube, to allow a rotation without re-clipping.
- The return of locators in device coordinates was argued on the grounds that input and output are orthogonal and, therefore, there was no connection between output positions of the latest window/viewport mapping and the equivalent input position.
It was a long day starting at 08.30 and finishing at 22.30. Jim Michener was the main person who defended the Core although Bert Herzog, Bob Dunn and Robin Williams were also present.
The last day started with a detailed discussion of the major points of issue that came out on the previous day. Some of the items were:
- There was a large number of people who would support some method of pseudo device and file storage within the Core. Some support even for one which allowed bringing the information back at a different level from where it was defined, for example Clip and Store and then bring it back for Re-clipping.
- There was some demand for a well-structured naming within segments, that is BEGIN and END name primitives. The argument against was that PICK-ID was just an attribute like any other and so you would also need Structured attributes like broken. The consensus seemed to be saying that the best way out is not to call it a two-level naming system as it was not. Really two separate levels for different purposes and different effects is what was in the Core.
- Image Transformations were disliked by many but there was also a need to support high quality displays and the drag facility available in a number of systems. This also was regarded as an image transformation. Again it was decided to look at it again and try and come up with a way whereby these transformations could be omitted from the Core and yet still have the possibility of optimising a Core implementation to allow the hardware to be used.
- There was a lot of support both for and against locator returning results in world coordinates. Arguments against are that with multiple window/viewport transformations in use, it was difficult to decide which world coordinates to use. Sometimes you wish to locate and draw figures even though you have not yet set up a window/viewport mapping. There was also a suggestion that a similar problem existed with PICK-ID when using a tablet if you had multiple viewports.
Many other topics came up but I think these were the main ones of interest to us.
The rest of the day was spent in procedural matters. The main items of interest here being:
- Agreement by ANSI and DIN (Germany) to try and work jointly to a Core definition.
- The formation of a European group under Paul ten Hagen of Holland to improve the input side of the Core proposal.
- The Chairman, David Fisher of BSI, was replaced by Paul ten Hagen.
- The next meeting will be in September 1978 in Bologne. The aim would be to push ahead in the meantime. The ISO Group would be widened by allowing all member nations to nominate experts if they wish. Hopefully, some part of the Core would be in a state for putting before the next meeting.
- There needed to be a focus for the British efforts and this will be through BSI. The aim is to try and produce a detailed specification of a Level 0/1 Passive Output-only Core System.
The results of the meeting were rather mixed. Too much time was wasted on small points due to poor Chairmanship. Bob Dunn of the USA took over at one stage and the improvement was quite dramatic.
There was certainly a lot of input to GSPC which should have an effect on the way the Core is progressing.
The formal side of the Working Group is now much better organised or will be and this will stop a certain amount of political problems that have been arising. There seems a good chance of cooperation between DIN and ANSI which, again, will be beneficial. The German Group are much stronger on the formal side and take into account the plotter world more.
The general view of the meeting was that progress was still being made and people are hopeful of some firm proposal coming forward in the next two years.
4. IFIP 77
The conference was, as always, well attended with several 1000 delegates and up to 6 parallel sessions as well as a separate conference on medical computing. The conference tends to be unique in terms of the number of countries being represented. There were papers given by authors from 32 countries. It was interesting to see the number of delegates from Japan which has certainly increased over the last few years. There were 42 Japanese authors of papers and after USA, France, UK and Canada, Japan was the most frequent in terms of number of papers.
Most of the names in Computer Science were present including Dijkstra, Boehm, Bolliet, Brooks, Codd, Denning, Ershov, Foley, Gries, ...
The only real fault of the conference was the absence of a delegate list which meant you found people by accident rather than design. The other fault outside the control of the Conference was the Canadian Air Traffic Controller's Strike which meant that the first day was disrupted with several papers having to be rescheduled due to the authors being still en route by bus or train (and it takes quite a while from Vancouver).
The conference was split between two sites but at least they were quite close and could be reached from each other by underground walkways. A necessity when it rains most of the time.
The Exhibition was not large by NCC standards but had most of the mainframe manufacturers in attendance. There were only two good graphics systems on show - both video, by Aydin and Norpak. Neither company has outlets in the UK but would like to start. The systems start around $10K and can be purchased in a variety of configurations. I have a full set of information. The Norpak system has been used by Herb Bown and it now uses his IMAGE system as software under a proprietary name.
There were several Diablo-like devices from various manufacturers including a fast Japanese version which I have details of. About twice the speed of the Diablo and similar type quality.
IMLAC was showing the PDS-4 but had nothing new on the stand. Tektronix had their complete range on show but nothing new. Similarly for DEC.
On the whole, the Exhibition was disappointing. The only item of note was the number of Japanese OEM disc drives etc around. There is some evidence that they are going to have some effect on the market soon.
One afterthought - one company had a Parallel Processor box a la DAP on their stand and available to plug into your computer. I have details.
Cliff and I tried to cover as much of the Conference as possible and (at the time of writing this - Wednesday evening) have so far not been in the same session. The Proceedings is very large and contains all the papers so will not go into any of them in detail. Instead, I will just point out major items.
The first day had an opening invited paper by Professor Wilkes that Cliff attended. The general impression I got from talking to people was that it was full of platitudes and not very good. He was supposed to be talking about likely advances.
The Program Management side seemed to be coming up with even more jargon. Having done structured programming, chief programmer teams, program validation etc, they were making the point that the software was no more reliable but more expensive and took longer to produce. The latest in-phrases seemed to be Requirements Specification and software metrics. That is the software was not working because they had built the wrong thing! They now have whole languages in which the user can specify rigorously (?) what he wants the software to do. They then generate a number of metrics so that the cost can be estimated and resources required allocated. It all seemed nebulous. A book that got mentioned several times was called Software Metrics by Gilb.
There were a lot of papers on CAD and Communications. I did the former and Cliff did the latter.
The most impressive of the CAD systems was one by a Japanese called Hosaka which produced complex objects starting from a solid block and cutting and slicing until the desired object took shape. The system is written in FORTRAN, runs on a Tektronix 4014, and can produce either hidden line or grey scale views of the object. The PLOT14 package is used. Input is via the Tektronix tablet and includes hand drawn characters as well as location information.
The results obtained were excellent. Apparently Leicester have a copy of the program but it is too large to run on their PDP11. It would certainly be worth getting hold of as it would be a valuable tool for some of our users. Time to recognise hand written characters was about 10 secs on a Prime 400-size mainframe.
Another paper of interest was from IBM San Jose and concerned the automatic overlaying of existing graphics with computer generated. The system was based on an IBM 5100 with a colour video system attached. It had been found to be of use in map making, engineering drawings, office management, etc. There was a great deal of cost evaluation and assessment to show that it was viable economically and was justified in decreasing errors etc. A fuller description is also available in a recent edition of IBM System Journal. Figures were like: operators found it 4 to 7 times faster, 2 to 4 times less errors, etc. Paper was by Carlson, Williams et al.
the system uses a Conrak TV Monitor and an Applicon Ink Jet plotter (that is the Swedish one made under licence).
Although you would probably call the work of Negroponte's group at MIT as CAD, I treat it separately as one bares virtually no resemblance to the other. The paper in the Proceedings (written about a year ago) was rather pessimistic about the future of CAD. His talk on the other hand was most optimistic.
He made a number of points about interfacing to a CAD system including a strong statement that colour was absolutely essential and suggested that input using the fingers was an area that needed further work. He showed the use of fingers to hit light buttons and move items on the screen including the use of more than one finger to get hold of several items together.
However, the real difference came when he talked about their current system and made the point that the computer system should not be the place you worked at but, in some sense, should be the total environment.
They currently have a single-user dedicated room, room is a better definition than computer system. On one wall is an 8ft by 12ft colour video display. Sound is octaphonic around the corners of the room and is linked in to the system and can output data in the same way as the video display. The user, or should I say occupant, has a leather chair which has data pads in the arms, a leather bound tablet which you can rest on your lap. The room has four inches of sound absorbing foam around it so that it is very quiet. Data can be stored with visual and sound cues. You can organise your data so that it is more like a library, ie on the top left hand corner somewhere. It is possible for you to play back a whole set of data (noise at a cocktail party type) and pick out the relevant one as a means of data extraction.
The screen will eventually be touch sensitive. Sumagraphics are just building it.
The only problem at the moment is that the system needs four dedicated Interdata 7/32s to run it with 1.5Mbyte of main memory.
There was a lot of discussion about Standards with the audience, on the whole, being aware of what was going on for the first time. In general, the view here, as at SIGGRAPH, was that, although storage tubes and refresh displays will be around for a long time, video colour displays would be dominant in a few years.
Probably the most effective paper was by Fred Brooks (Mythical Man Month fame) about the research being done at North Carolina with Jim Foley et al. He made a good case that Computer Science was an engineering rather than scientific discipline in as far as we were making things, even if they were algorithms, rather than looking for fundamental laws. He made the point that computer graphics was like toolmaking and the important point always must be the quality of the tool to do the job and this could only be done by seeing their effectiveness in real user applications.
He described a number of experiments they have conducted on PhD theses to see the effectiveness of graphics and these all proved statistically significant even when the graphics was simply used as a fast page turning mechanism. The ability to keep your train of thought going was important.
Several of the projects involved input using tactile output stimuli to enhance the visual effect. Manipulation on the screen was done using what looked like a remote sensor device which increased or decreased resistance. For example, getting hold of an object you actually felt yourself pick it up and could distinguish between different types. Such tactile cues were shown to be significant.
One area of interest was molecular rotation where they have had 13 groups using the system for 1500 hours analysing molecules. The figures suggest that the work is speeded up by a factor between 4 and 10.
Full details with additional references are given in his paper.
There was not a lot that was new here. There were a number of papers on Compiling but these tended to be rather specialised. A couple of TWS were described including quite a good one called MUG2 from Germany (paper by Ganzinger).
Ken Jackson of RRE talked about two languages MORAL and MASCAL for real time applications. The first is rather CORAL like but is more in terms of cooperating parallel processes than the CORAL block structure. That is, for example, data tended to have keys attached and the only processes that could access it had to have the key. MASCOT is a documentation system around MORAL.
He is a good speaker and I will try and get him invited down this autumn.
There was a Japanese paper by Ikeda on Preprocessors and a language which had a few special constructs but otherwise could be used to generate either FORTRAN or COBOL. Seemed similar to FOREST but there were a number of differences including what looked like a fairly comprehensive macro facility. They didn't allow GOTOs as all.
He estimated that productivity had increased by 30% and the users enjoyed using it.
I met Steve Levine from Livermore at the Conference and he was still extolling the virtues of their Dicomed D48 over the FR80. The machine is basically a point plotter so that there is no real vector start-up time. Long vectors are, therefore, considerably slower than the FR80 but short vectors are drawn much faster. He was quoting figures of 4 times the FR80 in speed on complex colour frames. I have a complex molecule slide which took 30 secs. Colour output is better than the FR80 and the system has one improvement in that the intensity range is automatically set up separately for each filter, that is the 1 to 256 define different intensities on each filter and so you can do the colour compensation there. Allows you a much better chance of matching intensities. To allow them to continue their colour development, Livermore have also purchased an FR80 colour system for production work! They use FR80 Displayer commands on both. The Dicomed costs about $120K and you have to interface it to a specific computer, eg PDP11.
One interesting fact: Litton are going to produce tubes with a mixed P45/P22 Red LP216 Phosphor, which gives similar granularity and about 2 to 3 times the brightness of the current white phosphor tubes. It was not obvious whether III were aware of this although I am sure they must be.
The Dicomed slide I have was done at F5.6, Slow Film, 1 HIT so their system has more light output than our FR80 in any case.
on the whole, the Conference was well worth going to. The papers and audience were both very international. I think it is still the only Conference having this flavour. The USA was well represented but so also were many others.
The UK scene did not stand out too well. The paper by Wilkes was particularly bad. Quite a few papers around MU5 and the Manchester University scene but the quality was not high.
Many of the continental papers were well received especially in the more formal areas. I heard good papers from France, Hungary and Germany.
The Canadians were more in evidence than usual and seem to be doing good work on both networking and graphics.
The major point was the great involvement of Japan in terms of speakers, attendees and exhibitors.
Cliff and I flew down to Prime at Framingham on Friday evening so that we could see Bill Poduska, Bob Murrell and Ken Fisher (President) on Saturday before they left for a Board Meeting somewhere in the Bahamas!
Most of the meeting was spent in us giving them information including a description of SRC's activities. We made the point quite hard that we saw that Prime had a special position in the UK being recognised more than they were in the USA and that there was a good reason for more active support of the UK Company including possible investments such as doing software development in the UK. They apparently are considering this and are taking it seriously. Youden, one of their systems people, will be staying in England for an extended period to give the UK Office additional support.
The meeting was particularly valuable in that on the full day Monday meeting we were able to bring up a number of topics raised by Bill Poduska on the Saturday. In some cases, I don't think we would have got quite such full replies without the people knowing that Poduska had offered the information.
I will add the information gained on Saturday in the Monday comments at the relevant places.
We had Sunday off and Cliff Pavelin, Bob Hopgood, and Patrick de Payer from Prime drove down to Cape Cod and Bedford to do some sightseeing.
5.1 Introduction (Bob Claussen)
Bob Claussen, who is on the marketing side of the organisation, introduced us to Prime as a company. He indicated that the company had made a number of basic rules at the start of its life. The first was not to make peripherals with its high R&D requirement. He mentioned that they do 10 cents worth of testing on each IC chip. Early on, they decided to concentrate on MOS memory and had never delivered a system with core memory.
The financial situation of the company was good. They had just been able to raise their bank credit limit from $10M to $20M funded by a set of banks including Citibank. They aimed for a 20% Profit Before Tax and would realise between 15% and 18% this year.
They have recently decided to standardise for VDUs on the Fox and Owl terminals of Perkins Elmer. One is a buffered VDU.
The organisation map within the company is given below:
As well as standardising on VDUs, there seemed to be a likely standardisation on Vector General for refresh displays and raster scan from Ramtek. The latter is very tentative and we voiced our disapproval on cost grounds. We made a similar point about Vector General and also brought up the point about maintenance in the UK. There seemed less display experience in Prime than there is amongst the UK Prime Users.
5.2 Visits (Dave Johnson)
This was mainly with regard to Cliff's visits, a brief overview of the sites that he was to visit were described.
They have 5 or 6 Prime systems. Some are general time sharing systems although the main emphasis is in the Medical School running MUMPS which they converted from a DEC product.
The main emphasis here was business administration and management science. They were the first customer to install SPSS. They were also involved in critical path analysis, simulation of management decision processes. They ran 30 terminals on their Prime system.
Near Albany, a university which has an Amdahl mainframe and had purchased two Prime 550 systems after an extensive evaluation. They happen to have a number of older students in places of influence in some large companies and were able to use these (three General Motors Vice Presidents). General Motors, as a result, did a technical evaluation as well as a financial evaluation of Prime's stability. They were eventually impressed by the architecture. As a result they may now get orders from General Motors.
The system here consists of two Prime 500s each with 1 Mbyte of memory and two 300 Mbytes drives. They expect to support 30 users on each system. Application areas include Finite Elements, ICES, GENESYS, SPICE, Circuit Analysis, Logic Simulation, Graphics etc. They are using 4014s and Imlacs.
They were using the software developed at Harry Diamond Labs, GRAFHELP, and there were other developments in terms of graphics software which were sponsored by General Motors and Prime would have access to these results.
There was some discussion about the relative goodness of Vector General and Imlac. Prime seemed pleased with Vector General mainly on the experience of SCI who had developed a circuit design package which, due to the power of the Prime, could often get 98% routing efficiency compared with 50-60% on Nova and DEC equipment.
They were likely to go for VG3300 rather than 3400 systems. Only 20 of the latter systems have been sold this year.
5.3 Factory Tour
There is quite a lot of spare room at the Prime plant now that some of the people have moved to a new building at Wellesley. General impression is that there is a very friendly atmosphere, staff enjoy working there and they have no staff problems. Quite small compared with DEC. Only 85 people are involved in the assembly and testing of systems. They pride themselves on having only a 7% turn over of staff per year. There are about 30 Prime systems on the floor being assembled/tested at any one time.
The new systems are orange (possibly Hot Tango!) and have a new stand-out display panel still with keys. I prefer the old. The cabinets are wider and shorter. The power unit (now made by Prime) is at the bottom and cooling is forced up the side of the cabinet. Thus, apart from the side cooling, width is as now and our logic could be put in a new cabinet.
The shop floor was filled with 80 Mbyte and 300 Mbyte drives (they wouldn't let us take one). The latter look similar but with more surfaces on the disc pack. They have had no problems anywhere with the 80 Mbyte drives and it seems as though we are unlucky and have rogue drives. Cliff may find out more on Friday.
The 300 Mbyte drives were introduced in June last year and were available by January.
5.4 Documentation (Steve Andelman)
Prime, thank goodness, finally seem to have seen the light. They have now set up a whole department to solve the documentation problem. They now have 8 technical writers, 1 senior editor, 4 technical typists and have increased the budget of the Department to $300K. They hope to bring out a number of new manuals by Rev 14 (due November 1) and hope to virtually have a complete new set by Rev 15 (6 months later).
They have organised project teams involved with each manual with representatives from the various sides of the company as well as users. The main manual in development or, at least, late development is the FORTRAN User's Guide which is a complete rewrite. We saw the current state of this. It looks far better and describes all the new parameters to the FORTRAN command (in English!). They have agreed to let us be one of the commenting list of active participants. Should be available in preliminary form by Rev 14. Each manual is likely to go through an IDR (Initial Document Release) stage - often produced by RUNOFF before it goes to the FDR (Final ..) stage. The intention is to use OCR techniques to input the pages to the photo typesetter. Amendments will be made by Change Sheet Update.
They implied that there were facilities in RUNOFF that we didn't know about including automatic page and section referencing. This will be in the new RUNOFF manual available in preliminary form by Rev 14.
There is a complete reorganisation of manuals and some will disappear completely. The intention, for example, is that the FORTRAN Manual should nearly be self-sufficient including LOGIN, file store, editing etc.
Unfortunately, it currently documents LOAD before SEG etc and we had a moan about that. The Development Systems Manual should be obsolete by Rev15.
A P400 Assembly Language/Reference Manual should be available by Rev 14. General impression is that we should see a significant change at last.
5.5 Hardware Product Plans (Tom Doyle)
in the CPU area, the PROM space should increase. They have a 4K PROM in-house and will have an 8K PROM by the end of the year.
They do not see any immediate need for a true 32-bit machine but will look at the market.
In memory, they currently have the 256 Kbyte board and have been looking at bubble memories etc as an intermediate storage medium. They do not expect a 64K chip (1 Mbyte board) until late 1978. The bubble memory work is at least a year away.
The 300 Mbyte drives will be enhanced by CDC to 600 Mbyte drives next year. This will be 2 spindles in a single enclosure. As yet, they cannot get double the density on the discs.
They are looking at Winchester devices for intermediate storage. These are sealed units having a capacity in multiples of 12 Mbytes up to 96 Mbytes. Non-removable, very reliable and fast access. Estimate mean time between faults is 5000 hours. Highly reliable and 25% less cost than cartridge discs.
Tapes are likely to include a 6250 bpi, 125 ips drive. Needed for back-up. At the moment, a complete disc copy of a 300 Mbyte disc to another disc takes 1 hour and 20 minutes.
There were two main advances in the Communications area. The first was ELF (Extended Logic File). This project proved much more expensive than expected and, consequently, would never be a standard product. Two customers will be getting specials (Hawker-Siddley is one). It had been hoped that it would only be 1.5 times the cost of AMLC. Currently they are reworking it to get the same performance at lower cost. It has auto speed detect in microcode. A driver does exist and will be finished for delivery in October. Hawker-Siddley will use it to run 4014s at high speed. The other customer is Rennselear. The new ELF product is expected some time in 1978.
Work is being done on an X25 chip and will be available by rev 15.
In terms of communication, PRIMENET appears to be totally different from DECNET. The lower level protocol is identical to TELNET.
5.7 Rev 14/15 (Mike Greata)
This was a long session which started with a description of possible inclusions and finished with Cliff going through in detail all the questions we had accumulated.
The main new items in rev 14 will be 64 V Mode COBOL, RPG and DBMS. PRIMOS 4 will use process exchange with each user an independent process. This means that the mini scheduler is no longer required. The scheduler has been rewritten so that most of the WAITs are on semaphores. It is possible to examine the semaphores to find out lengths of queues.
Internally, there will be fast user swapping, lower system overheads and increased overlapping of disc I/O and other activities. Throughput should be up to Rev 11 level as long as you have adequate storage. Rev 14 is a speed release. Rev 15 will aim to reduce store requirements.
Priority levels and time slice per user have been added. Possible to have inter-communicating processes. An operator command exists to allow phantoms to be run at a lower priority level.
Currently, Rev 14 is being used in-house. Snooper program shows that there is a 75% overlap of user program and disc access (Cliff thinks we have about 10% at the moment).
It is likely that Rev 15 will have optimisation of disc seeks so that disc will be swept for files. This is likely to give even better performance if disc is cleaned up from time to time.
Rev 14 is likely to boost the paging rate as it will let more users get work done.
Rev 14 will have the new parameter system with English rather than Octal parameters. It appears as though these can be in two forms:
I = LIST .. , -LIST, ..
We didn't get many details but they appear in the FORTRAN Manual which Cliff is getting a copy of on Friday.
Improvements have been made to the AMLC code. Mainly simple - ie don't do work when it is not necessary. Overhead may go down by a factor of 2.
Files will be able to specify number of readers and writers on a file basis in Rev 15. In Rev 15 it will be possible for a user to get at Ring 1 t protect parts of a system. Rev 14 has a Ring 0 for each process. The stack for Ring 0 is locked in memory so at least one page has to be locked down per configured user.
The three main dates for the Rev 14 release are:
- A: Documentation to the Technical Writers (Yesterday)
- B: Code finished (mid-September)
- C: In field by November 1 plus a few days
Likely to be about 6 weeks later in the UK.
Rev 15 will define max and min storage allowed for UDFD which is set when the UDFD is created. There will be an operator command to specify maximum number of users.
Routines will be available in the Applications Library to allow unique temporary files to be created, opened and closed in a directory (Rev 14).
Emulators for RJEs will e cleaned up in Rev 14. Likely to extend time slice to guarantee response. Will make our system worse. Documentation for 2780 will be available. There appears to be no plans for enhancements to the 2780 or HASP in the way Graham has already altered it as far as enquiry and return files to correct filestore.
Rev 14 should support the synchronisation communication between Prime 400s, Process-Process communication and file accessing on a remote machine. Compile from a file on another machine is likely to be about 6-7 times longer.
By Rev 15 the intention is to have the ability to LOGIN to a remote host.
A loop network facility called FARRNET is being developed for close coaxial cable connections running at 8 Mbits.
Debugging of FORTRAN programs is being considered. Retroactive trace and array bound checking etc. Likely that Symbol Table formats will change. May see something by Rev 15.
The time allowed before carrier stabilisation has been changed to 0.2 sec (currently set at 0.1 sec). Should help to cure collaborator login problems.
There appears to be a mechanism for forcing the logout of a user.
Filestore back-up on an incremental basis by file or disc will be available by Rev 15. Both will be available by Rev 16.
Prime have a program for checking library order and will make it available to us. John Howell of Prime, Bedford apparently has some knowledge of Tektronix tablets.
These are a fairly random set of comments. Cliff will provide a more thorough document on his return. It was a good day and I think we have a much better idea now in which areas Prime are likely to work. I have a list of new manuals and their status.
I visited IMLAC on Tuesday and talked to John Fadiman, Leonard Sacon and Ray Fortin. I left to travel back to the UK that evening. Bob Slater of Prime was interested in seeing the facility and so spent the morning there also. He left after lunch and IMLAC then discussed their possible future plans with me.
IMLAC are considerably smaller than Prime but are a similar company in many ways. They were formed about 9 years ago by a number of business men who wanted to invest in an interactive display production for stock exchange use. They decided to form a company to manufacture it.
By the time the product (PDS-1) was finished, it was obvious that the idea was not really on and the managing director left. Leonard Sacon arrived at that time as head of marketing and was left with a product and no market. It was his decision to go into the graphic display area and market the PDS-1 there.
the company works mainly as an assembly organisation not making anything themselves. The factory at the moment is in complete chaos as they are reorganising everything. PDS-4 were lying around everywhere! They tend to produce standard items for stock and then take them from stock to build a complete system. They had 20 to 30 PDS-4s in their stock room.
It was quite obvious that they had a lot of orders and did not start a system until about 6 weeks before delivery. The Reading system is due for shipment in mid-November.
It was interesting to see the software development section each with its own PDS-4!
Documentation is produced using a small in-house printing system with most of the output being typed on a Diablo. All manuals are kept in source on an IMLAC system.
The USA company appeared to have little knowledge or interest in DINO. They marketed or at least provide two systems. The first was GRAFHELP (Harry Diamond Labs) which we have the source of. This is quite popular and has an IMLAC resident part GTS which is about 4K in size.
They also provide GCS with an IMLAC resident part called TIS which needs about 8K. It allows multi-level segmentation and can be used in stand-alone mode. I had a demonstration and it is quite flexible allowing segment copying etc. Nesting is allowed up to 128 levels.
As a company, IMLAC is leaning towards GRAFHELP as a supported product. They feel that the high USA cost of GINO makes it impossible.
I have a copy of the TIS specification. We can get the binary free of charge and the source for a handling charge. Could be a possibility as a GPGS-F handler.
IMLAC sales in the USA have been good. There are 35 PDS systems in the UK of which 25 are PDS-4s. There are 700 installed in the USA. They manufacture 3-4 systems a week.
They decided last year to start a subsidiary in Europe and, because of low sales, had decided on France. Unbeknown to IMLAC, Cetec organised the merger with Ferranti and that forced them to change their plans. They seem very happy with the result.
They have a cross-assembler produced by LOGICON for the Nova. It is written in Data General FORTRAN and is available. I have a listing. It may be of use to us.
They are working on a new product and I have details although the UK company have not been told yet. It is confidential but they regard us as an influential customer having 3 out of 700 PDS-4s! I will tell the people who need to know.
I raised the problem we were having with the tablet. The standard method of solving it in the USA has been to ignore the lowest bits in terms of accuracy.
It was interesting to have a demo in the factory with a SAC tablet and no problems! the pens apparently wear out but otherwise no problems.
On the whole it was a good trip. The days at Prime and IMLAC were very useful. IFIP was a bit spoiled by the traffic controller's strike and there were fewer delegates than they expected. However, there were a lot of ideas especially in the graphics area.
As for graphics standards, progress is being made and although not much was achieved at this meeting, a good deal of procedural matter was sorted out which will make things easier in the future.
This plane is still going up and down. It must be the worst Atlantic crossing I have had!