A large part of 1983 involved Colin Prosser in starting to assess the market for an alternative to the PERQ in the single user system Common Base.
In January 1983 a full set of benchmarks between the PERQ and Apollo system was carried out. The major conclusions were:-
Arithmetic Operations: integer operations on the PERQ were significantly faster than the Apollo frequently 4 times the Apollo speed whether using FORTRAN or PASCAL.
Real operations were more variable with single precision real divide on the Apollo being 25% faster than the PERQ while addition and subtraction was 5 times worse.
On the whole, the PERQ was a significantly better scientific engine.
The large variability made the comparisons on real programs difficult especially as both companies were making significant changes to both software and hardware. There was certainly no case in January 1983 to include Apollo in the Common Base.
As a result, the much wider survey was undertaken.
By March 1983, over 100 companies had been approached with the Operational Requirement put together at the end of 1982. The overall impression from the early replies was that despite the delays, the PERQ still had few real competitors. About 50 companies indicated that they could meet the OR and considerable time and effort were used in sorting through the information provided to see whether they were competitive. As many products were in an early state, the only way to do any real evaluation was to visit the USA and benchmark the systems. This was done for the leading contenders during 1983.
Eventually over 120 manufacturers were contacted. In some cases, it took several months for companies to reply so that identification of suitable products proceeded much more slowly than initially anticipated.
The number of real candidates was reduced to manageable proportions by removing those that:
This narrowed the field to 21 possible suppliers. Some of these could be assessed by telephone or by local contact. The visit to the USA in July did more stringent benchmarking and had further discussions with Apollo, Symbolics, SUN, Hewlett Packard and Xerox. There was a considerable amount of discussion with suppliers to explain the Operational Requirement in detail.
Notable manufacturers who refused to quote or soon withdrew were IBM, GEC. DEC, Apple, PRIME and even ICL. Also the visit to Xerox showed that they would not offer a UNIX system. Tektronix indicated that they had nothing to offer at that stage but might have in the future.
Low and Mid Range Products (£8K to £llK) were assessed and tended to fall below the requirements in terms of UNIX environment, standard bus, processor speed, floating point performance, virtual memory provision, fast discs and UK distribution and service. The most attractive system came from CIFER. Forward Technology, Valid Logic, Digital Computer, Acorn, National Semi Conductors, P1essey, Perkin Elmer, Pixel, Ridge and Whitechapel were all considered. At that time, Whitechapel were in an early stage of development and although recognised as having an interesting product were clearly not ready for assessment.
Of the High Performance Systems, the main comments were:
Hewlett Packard: the HP9000 had been announced with its small 560 × 455 screen. HP had put a great deal of effort into porting UNIX using two teams one of which did a conventional port while the other attempted a more ambitious distributed UNIX system (reminiscent of the ICL/RAL PERQ activities). In this case, the proprietary UNIX product proved superior and the conventional UNIX port was abandoned.
Both the technology of the product with its own purpose built processors and the quality of the UNIX port were impressive. They had put 30 man years of effort into the UNIX port.
The major drawback was that a system of comparable power to the PERQ was approximately twice the price. The system was also in a very early stage of development.
SUN: SUN had moved into a new 45,000 sq ft facility in December 1982, and were beginning to be a potential threat to ICL and Apollo by mid-1983. By June 1983, they were shipping $2.5M of equipment per month and, even so, the delivery time of 60 days had gone out to 90 days.
SUN had taken the same approach as Apollo in getting volume sales up early on. They had focused on OEMs and experienced customers likely to purchase in bulk and require little support.
The UK distributor Keen had been not very successful and throughout 1983 SUN were nervous to set up in Europe in strength until the USA market was sorted out.
The Company was still relatively small with only 125 people, much smaller than Apollo at that stage and about the same size as Three Rivers.
Certainly, if ICL had put their mind to it, their manufacturing capability would have dwarfed all three with no noticeable impression on ICL's total manufacturing capability. Such was the difference in size between the four main companies in the field.
The SUN systems at that time were not very impressive. They ran a special variant of Berkeley 4.1 UNIX and the discless nodes each had a separate partition of the file server so that true sharing of files was not available. The system had limitations similar to the PERQ on what the power supply could handle. If you had a 2 Mbyte system, it could not run on the ethernet nor could it be a colour system.
At that time, SUN were showing no interest in X25 wide area network communications. The window manager was very restrictive only 16 windows maximum, but the machine was not fast enough to handle these, limitations on the number of fonts available, you could not output to partially obscured windows, cursor tracking was very slow and there were bugs in the system. However much the user base complained about the PERQ window manager, the offerings from SUN, Apollo and HP were significantly worse.
A major advantage that both Apollo and SUN had over ICL was the colour system. Apollo had an integrated system while SUN had an attached colour monitor. Three Rivers were working on a similar product. ICL did not take up this option until mid-1985 despite it being available from Three Rivers in 1983.
Finally, the price of SUN systems, which were being praised by the UK community as very cheap, turned out to be expensive. All the software was unbundled and by the time you had a reasonable configuration, the cost of a SUN1 was about twice the cost of a PERQ with lower performance.
The conclusions from the assessment done in 1983 were that, at that time, there was no single machine substantially better than the PERQ in every respect. The closest competitors (SUN, APOLLO, and Hewlett Packard) were approximately twice the price for equivalent physical configurations.
The PERQ was particularly strong in terms of interactive graphics capabilities. The weakest features of the PERQ were in the areas of virtual memory management and inter process communication. ICL had plans with PERQ2 to remedy both.
It was clear that the market place, by the end of 1983, was still lagging behind the expectations of the users but that interesting developments were underway by the companies mentioned above and a host of others including large companies like DEC and IBM.
The RAL group recommended to SUSSG that no immediate action should be taken to purchase alternative single user systems to the PERQ. Further market developments should be awaited and only once a clearly equivalent or better machine was identified should this position be reviewed.
The SUSSG broadly agreed with the conclusions. They had a particular problem in that funds were not available to support a second system even if it was recommended. There was concern that the price of the systems was going up rather than down which meant that putting a large number of systems in one location was prohibitively expensive and, consequently, little research was going into the effect a complete environment of single user systems would have on the research community. It was agreed that RAL should look at what could be done using the IBM PC as the de facto standard in the lower cost range. [As a result of this request, a group was set up to look at the IBM PC and this continues in Central Computing Division at RAL today funded from the central infrastructure budget and providing guidance on IBM PC matters.]
The SUSSG decided to recommend no immediate action should be taken to purchase alternative single user systems to the PERQ.
It was agreed that when the new products from companies became available in 1984, a full evaluation of these should take place with the possible purchase of a number of systems for detailed assessment. The aim would be to have a second system in the Common Base by the end of 1984 if one was available on the market by then.
