ACD Atlas Computing Division PERQ History: Part VIII

Jump To Main Content

Jump Over Banner

Home

ACDSingle User SystemsPERQ HistoryPart VIII

Jump Over Left Menu

1984/5

47. HARDWARE

47.1 Introduction

By March 1984, the only system being delivered by ICL was the PERQ2. Initially these were with the A4 displays but very soon larger landscape displays became available. ICL saw a 1 or 2 Mbyte system with 16K WCS and landscape display as their main system as far as marketing was concerned and future software developments were based on this as the standard configuration.

The first pre-production batch of 50 PERQ2s had been manufactured near the end of 1983 and full production systems started to appear around March 1984.

47.2 PNX 2.0 Hardware Upgrade

To move from PNX1.5 to PNX2.0, a hardware upgrade to both the cpu and i/o board was needed on all PERQ1s. This cured some known faults and changed the ROM in the Z80 I/O processor. Early in the year, it was agreed that ICL would systematically go through each Region in turn upgrading to the new hardware level.

Each system required an upgrade kit and the aim was for engineers to do a complete location in one go, having first contacted the site the week before. There were 142 systems to be done, with an equal split between ICL's four main support regions. The initial aim was to complete the operation by the end of June. There was a significant amount of work to be done and, although not happy with the schedule, it was probably a realistic one.

There were immediate problems with the upgrade programme. By April 1984, ICL had not started the upgrade programme apart from 3 systems that had been upgraded on an ad hoc basis. ICL, at the beginning of April, wished to put the whole programme back by two months, with a completion date being the end of August.

[It was much later on we found out that ICL had posted upgrade kits from one ICL site to another and a large number had been lost in the post!]

There was a series of quite angry letters between SERC (Dr Manning, Prof McGreavy (Chairman of the SUSSG)) and ICL representatives, both at the Director level and to Charles Hughes, the manager of the PERQ Business Centre at Kidsgrove during April and May. There were also two meetings at Putney and RAL. It was indicated that there would be a strong adverse reaction from the user community to any slippage. There was already a credibility gap with ICL and this would become a chasm and might well jeopardise the whole single user system programme.

The SUSSG was becoming quite uneasy about the hardware and software progress from ICL. There was strong pressure to withdraw PERQ from the Common Base entirely until the deficiencies were removed. The complaints were general and widespread within the community and ICL could not ignore them.

ICL were in the middle of launching PERQ2 and this was not the best time to use a great deal of effort in retrofitting PERQ1 systems.

As a result of this flak from SERC, ICL did put effort into bringing forward the dates. SERC divided up the systems into Immediate, Very Urgent and Urgent. The Immediate ones would be upgraded before the end of May.

By mid-May, a revised programme which completed the Immediate ones by the end of May, the Very Urgent by mid-June and the Urgent by the end of June was defined. These lists formed 83 of the 141 systems to be upgraded. The lists were constructed on the basis of known requirements and activity from their contact with the Support Office. Users who had no hardware problems and were happy with existing software (the remaining 58) were not on these urgent lists.

ICL never achieved these dates and, eventually, the end of August date proved optimistic. Each system took longer to upgrade than anticipated, the loss of upgrade kits meant that fewer were available than required and the ICL support staff were not all familiar with the hardware. Consequently, some regions achieved greater progress than others.

By the beginning of September, 4 out of 10 of the Immediate systems in the North East/Scotland region had not been upgraded, while all the other regions were complete. Two regions had completed the Very Urgent and were working through the Urgent, while the North East/Scotland and North West/Midlands were quite a way behind.

Key staff going on holiday during this period also meant additional slippage once the holiday season arrived.

It was not until October that all systems on the three lists had been successfully upgraded. Once ICL senior management had become fully aware of the problem, great pressure was put on to the ICL engineers resulting in significant amounts of overtime and weekend working. An example is one engineer who had a nervous breakdown, was off work a few weeks, came back and promptly fell ill again (nobody had done any of his work during his absence!). Consequently, although PNX2. 0 had been available near the beginning of the year, most users were unable to take advantage of it until the end of 1984.

Another problem occurred with a small number of users who had based their application programs on an early release of the POS operating system that was incompatible with the latest release of POS. In the end, RAL had to ask ICL not to upgrade some systems.

We had also done a number of upgrades prior to the general release of the upgrade kits to allow field trials of PNX2. 0 to take place. The upgrades to these systems were non standard and we eventually had to refit these with the proper upgrade.

All of this was quite traumatic and, due to no decision on the 16K WCS upgrade, was going to have to be repeated if the 16K WCS upgrade was agreed later. At least the 16K WCS upgrade was only a board swap once these changes had been made and would be less of a problem. Even so, significant effort would have been saved if they could have been done together.

Thus, although a reasonable version of PNX with faster compilation was available in 1984, the only SERC users able to benefit from it were the PERQ2 systems. These were mainly EB grants and the Software Engineering side of Alvey.

47.3 The 16K WCS Upgrade

Section 44.3 gave the situation at the end of 1983. ICL were not prepared to run PNX2.0 and GKS on 4K WCS systems. The microcode needed for the two products was greater than 4K and future changes were likely to increase the microcode requirements. The performance of 16K WCS systems was projected to be significantly better than 4K WCS systems (predictions were 3 times on floating point performance and 10 times on GKS graphical output).

At the start of 1984 we had 132 4K WCS systems, of which 39 had been delivered with 16K WCS and needed to be paid for if kept. A bulk order of 40 systems near the end of 1983 together with some funds already available for 16K WCS systems allowed a deal to be organised with ICL whereby the 39 16K WCS systems were upgraded.

To upgrade the other 90 or so systems, ICL agreed early on in the year that, because of the special relationship with SERC, they would offer a 16K WCS upgrade as part of the PNX2.0 hardware changes at a cost of £lK per system instead of the normal cost of £3196.

The various funding bodies were asked if they would upgrade their systems at this cost and approval was obtained for most of the EB systems early in the summer with the hope that this could be done as part of the PNX2.0 upgrade. By June, commitments had been obtained for the upgrade from most of the funding bodies. However, over 6 months had elapsed from the original offer and ICL were now seeing that they would lose a considerable amount of money on the deal. Also, with threats of removing PERQ from the Common Base, it was less than clear that this was a reasonable thing for the PERQ Business Centre to subsidise.

There were strong letters from SUSSG's chairman to ICL in June 1984 indicating that unless the offer was reinstated, SERC would withdraw from the agreement with ICL and would not allow GKS or Spy to be used by ICL without payment.

A meeting was held between Dr Manning and Mike Watson of ICL on 25 September where a revised offer was made of £1400 per 16K WCS upgrade, a discount of over 50%. This offer was accepted by RAL on condition that ICL committed to replacing the current virtual memory system on the PERQ by an improved system as required by the AI community and some other users. There was a set of about 10 improvements put to ICL. Some proved impossible to do, such as replacing the power supplies. Many were accepted, as well as the virtual memory change. For example, ICL agreed to mount the new ERCC FORTRAN on PERQ1s. All together, the set of changes extended the PERQ1's life by possibly two years.

This commitment was obtained near the end of 1984 and by October 1985 all systems had received their 16K WCS upgrades apart from one or two.

47.4 Maintenance

A review was made during 1984 as to the effectiveness and cost of the maintenance provided by ICL. If the machines had been very reliable, there was the possibility of dropping the maintenance on the systems for a more cost-effective operation (board swap or replacement system).

In the period from January to November 1984, there were 295 engineering calls routed through RAL amounting to 1.5 per machine. Although this was the approved route, some users did make calls direct to ICL and the number of call-outs was likely to have been significantly higher.

The general view was that with the recent upgrades and future ones, there was much to be gained with having the block maintenance agreement. A further review was agreed in 1986.

47.5 PERQ3

In March 1984, ICL gave the first detailed presentation of PERQ3 to Prof McGreavy and other SERC representatives. The main points from the presentation were:

  1. PERQ3 would be a family of products with both low cost and high performance options.
  2. All systems would be 32-bit processors using a standard memory bus with rasterop functions closely integrated with the memory.
  3. All systems would have large landscape displays.
  4. The initial system would be PERQ3A, a low cost system with 2 Mbytes memory extending to 32 Mbytes. The machine would be very small in size. Cost would be half the current price (about £llK). This would probably be offered as a discless node.
  5. PERQ3B would be a disc system similar to PERQ2.
  6. Various accelerators would be provided to increase the basic performance of the system.
  7. ACCENT would be the standard operating system with UNIX built on top. Three Rivers were just releasing ACCENT on PERQ2 as the standard operating system, although ICL had decided to stay with PNX on PERQ2.

The overall concept of a fast system bus (44 Mbytes per sec to memory) with 48 bit virtual address allowing a number of processors to share the bus was exciting and allowed a readily extensible system. Of the 48 bits, 16 were used to specify the address space identifier and the other 32 bits per process.

The initial processor would be 68020 based with the top end processors being bit-slice technology like the PERQ2. The aim would be to start running as 12.5 MHz systems but aiming to run at 20 MHz for the bulk production version.

A major innovation was a 4096 × 3120 landscape display being jointly developed by ICL and Three Rivers. This was something unlikely to be available from competitors.

An environment with several PERQ systems would have ethernet based servers for disc, tape, printing and low cost peripherals (streamer tape, floppies etc).

The aim was to have the PERQ3A and PERQ3B available by the autumn of 1985.

Between them ICL and Three Rivers had about 30 people working directly on the product with the aim of increasing this number to 70 by the end of 1984.

SERC were impressed by the presentation, particularly the hardware. There were reservations on the software side. SERC had already been through the problem of getting a fast UNIX system based on ACCENT and could well see the system providing a good distributed system based on ACCENT but with poor UNIX performance.

A second meeting was held in June 1984 to discuss SERC and Alvey's comments on the initial proposal. At this second meeting, Colin Prosser represented ICL rather than SERC, having just moved to Kidsgrove from RAL.

The PERQ3A specification had become a lot firmer with a mock-up of the system in terms of boards and overall size being available. The PERQ3A system would initially be limited to 4 Mbytes with a possibility of 8 Mbytes later (using 256K chips). A fast cache pager board was being developed using high performance chips which would greatly boost the virtual memory performance.

The software situation was much more fluid with various options being looked at. There was a clear need to improve the virtual memory performance of the PERQ2s and a number of proposals were being looked at, some of which allowed the PERQ2 long term compatibility with PERQ3, assuming the ACCENT operating system was released on the PERQ3.

The main thrust at that time was to develop PNX to be more ACCENT compatible with the possibility of a move to ACCENT at a later stage.

As about half the effort on the project was coming from Three Rivers, there was a strong ACCENT lobby and this was clearly seen in the proposals at that stage. Everybody agreed that the conceptual design of ACCENT was the right way to go for a distributed set of high powered single user systems. It was a matter of how it could be obtained and keep compatibility with the existing PERQ2 user base. By now, ICL were firmly committed to not supporting PERQ1s even on PNX2.0.

A third meeting at the end of August 1985 gave further progress on PERQ3A. By now, the availability of ACCENT early on had receded and ICL were committed to releasing PERQ3A with System V UNIX and virtual memory extensions allowing compatibility with the existing PERQ2 user base.

The system would be a 1 or 2 Mbyte system initially with a local disc. It had been decided not to produce a disc1ess node as ICL was unconvinced that high performance could be obtained by paging over the ethernet. Having a minimal local disc with the virtual memory and cache would guarantee good performance superior to competitors. Distributed computing would be provided via the Newcastle Connection.

PERQ2 developments included an integrated colour system.

The PERQ3A system was to stay much as defined here for the next year with prototype systems being delivered to RAL in November 1985. That was almost no slippage on the original timescales for the project. With Three Rivers going close to bankruptcy in the summer of 1985, ability to market the PERQ3A world wide was in ICL's favour.