The major task at RAL during 1984 was the assessment of the new products available from companies in comparison with the PERQ2. Permission had been given by SUSSG near the end of 1983 to acquire systems for this purpose either by loan or purchase depending on what arrangements could be made with the manufacturers.
An operational requirement was sent to Apollo, SUN, Hewlett-Packard, Racal, Gould SEL, Whitechapel, ICL, DEC, Ridge and High Level Hardware early in 1984. The timescale envisaged at the start of the project was to get responses from manufacturers by the end of March 1984, select systems for detailed assessment and benchmark these during the period April to July, getting the user community involved in the assessment. The aim was to spend August and September producing the final evaluation with presentation to SUSSG in October and changes to the Common Base being announced in November 1984.
The Operational Requirement was divided into essential and desirable sections with the software Common Base in the essential class. It was clearly stated that machines were needed early for assessment and manufacturers should not propose systems that would be available in the future. The Operational Requirement was a team effort, written up by Colin Prosser, and was a detailed 16-page document requiring a well formatted reply by the Company.
Of the 10 manufacturers contacted, DEC declined to tender indicating that they would have no systems available in the timescale. Detailed responses were obtained from the other nine companies in March.
Of the systems offered, the Apollo DN460/660 and the SUN2 came the closest to meeting the Operational Requirement. Like the PERQ2, they could not offer all facilities immediately but would work well as stand alone systems. The Whitechapel machines were felt by the SUS team to be not mature enough, but Alvey asked for them to be included in the assessment and provided manpower for this to be done.
It was felt that the HP9000, Gould PS5000 and RIDGE 32 might be contenders at the end of the year but not at this stage.
There were some members of SUSSG who were urging that either Apollo or SUN should be immediately added to the Common Base just on the basis of the paper assessment. However, it was agreed that both SUN and Apollo systems should be obtained for evaluation and that manufacturers should be asked to loan machines in preference to purchasing. All Board representatives were asked to consider suitable evaluation procedures.
A second meeting of SUSSG in May 1984 agreed to add the Whitechapel system to the evaluation. It was a British system and, even though in a very early stage of development, was clearly targeted at this market. Apollo had agreed to loan two systems, a DN660 and a DN320 (until the DN550 could be made available). SUN agreed to loan a 2 Mbyte and 4 Mbyte system for evaluation. The latter system would be assessed at the AI Department in Edinburgh together with an Apollo system.
At this meeting and the following one in July, the four Board representatives were urged to give details of how each Board would do the assessment relevant to his area as there would be a need to arrange access to the evaluation machines.
An Interim Report was given to SUSSG in July. The only Apollo system that had arrived was the DN320. The DN660 was expected in the middle of July. The Mid Range Computational Node system from Apollo had been announced as the DN550 and the company had written indicating that this could not be supplied in the required timescale. There were significant negotiations with Apollo before it was possible to get the DN550 in a timescale useful to the evaluation.
The first SUN machine arrived at the end of June with the larger machine destined for Edinburgh due to arrive at the end of July. Both Whitechapel machines were destined to arrive mid July.
The inability of manufacturers to supply equipment had delayed the assessment by about 3 months.
Further slippage occurred. The second SUN did not arrive until the end of August and both Whitechapels were delayed until mid August. Consequently, no report was possible to the SUSSG meeting in September.
Despite attempts to get feedback, neither the Science Board or ASR Board made any comments on the evaluation procedure. Comments were received from the Engineering Board and Nuclear Physics. Engineering Board played a major part in the evaluation.
By the end of September, all the systems had been delivered including the Apollo DN550. The SUN systems and the Apollo DN660 proved to be reliable, but problems were experienced with the DN550, and the Whitechapel machine had to be replaced as it only worked intermittently.
The final assessment was presented to SUSSG at its meeting on 28 November 1984. The evaluation criteria agreed by SUSSG was that the items of major importance were:
The next set of items also important were:
There were many other criteria but this set was meant to be the dominant ones in any evaluation.
The major comparison was done between the DN550, SUN2/l20 and Whitechapel MG-1.
The conclusions were:
The system had sound basic hardware but did not significantly outperform the competition except in graphics performance, where it was comparable to the PERQ2 and much faster than the SUN or Whitechapel. The UNIX implementation had some ragged edges. The window manager could not support a modern graphics oriented interface, although it had some features of merit. There was more software available for the development programmer and much more for the end user. However, it was very expensive. The benchmark configuration cost £74K compared with the £30K for the SUN and just over £10K for the Whitechapel.
One of the major aspects of the Assessment was to take the SPY editor as a demanding application program and port it to each of the evaluation systems. In the case of the Apollo, the constraints placed on the application by the window manager were such that it would have taken at least 3 man months to port with no guarantee of reasonable performance. After a paper exercise to see what was needed, it was decided not to proceed with an implementation.
This system was surprisingly fast considering the price on basic operations, but was marred by poor graphics performance. It had the best UNIX implementation and an adequate window manager. The SPY port was achieved in a reasonable timescale but performance was significantly worse than on the PERQ2. There were few aids for program development, although some were expected in 1985. The applications software base was sizeable and growing, although significantly smaller than the Apollo.
The major point in favour of the SUN 2/120 was that it had no really bad features which could automatically rule it out.
The basic hardware was impressive, particularly at the price. However, there was almost no software available, not even a window manager. It would be of interest in 6 to 12 months but was not a viable system at this stage. The graphics performance turned out to be very poor, at least 20 times slower than the PERQ.
In general, the assessment was disappointing. About 1.5 man years of work was put in locally, spread over a number of people, and probably 6 man months more at Edinburgh and other user locations.
The recommendation to SUSSG was that each system had significant flaws and consequently none were suitable to add to the Common Base. Although the PERQ compilers were significantly slower than the Apollo and SUN systems, the graphics performance and window manager was still the best currently available in the UNIX environment. Only Apollo had superior debugging aids to the PERQ. All three were significantly worse than the PERQ in wide area network communications. Neither Apollo nor SUN were committed to ISO standard local area network protocols, and Apollo's proprietary local area network which would almost certainly lock users into a single supplier situation.
After much discussion, SUSSG decided that it would not be possible to continue with only one system in the Common Base. It reiterated its support for the ICL PERQ as a Common Base system and recommended that SUN be added to the Common Base alongside PERQ.
The recommendation to the new Director of Computing and Computing Facilities Sub Committee was as follows:
SUSSG recommends that the SUN2 be added to the Common Base for purposes requiring a mid-range workstation. It is estimated that the effort needed to support the SUN is 5 man years per year. The SUSSG recognises the need for a machine for CAD requiring high quality graphics and cpu power and that this should be the subject of a separate evaluation including applications software which would require an effort of 6 man months. SUSSG also notes the need for a low cost workstation as exemplified by the Whitechapel MG-1 machine which is at the present time not in a supportable state. The Steering Group recommended the Whitechapel development most strongly to DTI for support.
At the time of this assessment, Alvey had already approved a number of SUN systems but without provision for any support. The SUSSG felt that this might result in a major disservice to the research community.
A major problem was that CFSC did not have the manpower allocated to support SUN systems and unless this manpower could be provided, the SUSSG recommendation would not be implemented. RAL contacted the Alvey Director who agreed to provide the 5 my of effort for SUN support in the early years on the assumption that funding would eventually be provided from the Engineering Board. As a result, CFSC approved the SUSSG's recommendation on 7 January 1985 and it was agreed to issue a Press Release to that effect. Unfortunately, Alvey believed that this would be inappropriate as it showed preference to a USA manufacturer (even though Alvey were the ones providing the support and had purchased the systems!). As a result, the SUN announcement was not released and the early SUN systems were purchased by grant holders while Alvey attempted to sort out its strategy.
Alvey accepted the need to provide some support for Whitechapel and purchased 25 systems in early 1985. These systems were placed with people who could add to the base software available on the MG-1.
Finally, some steps were taken to carry on with the top end assessment and a Silicon Graphics system was purchased for assessment during 1985 as the most likely competitor to Apollo and SUN at the top end of the Single User System performance range.
