Mainframe replacement

Back to main page       Top

A more heavily political choice was that to replace the DEC-PDP10 by 1980/81. Already over seven years old it was heavily used by all groups. Competitive tendering lead to offers from IBM, ICL, Siemens, CII-Honeywell-Bull, and DEC.

PDP1070 processor 36-bit (1973), hardware paging, 256 kwords 36-bit memory

• 4 RP02/3 40 Mb disks, 1 Ampex compatible 40 Mb disk (removable/interchangeable packs)
• 4 800 bpi NRZ 9 track tape units
• 2 DECtapes (7 track 0.75 inch block replaceable tapes 2.5 Mb)
• 1 paper tape reader (boot device)
• DECwriter LA30 console, VT05 operators monitor
• 2 GT40 graphics
• 2 Tektronix 4010 graphics+hardcopy,
• Calcomp plotter
• 2 LP10 printers
• 10 user terminals - ASR33 teletypes 10cps
• 3 Network instruments (D11, D17, IN6) ILL protocol; serial line 9600baud)

The performance of the KI-1070 machine cpu was similar to the 1978 VAX11/780, though the disks were connected through channel hardware. (Later VAXes used proprietary interface, the BI bus interface chip, offering better performance than the basic UNIBUS).

New machine requirements :

• Approximately 3-4 times processor speed. Disk capacity 2 Gb
• Network front end processor/s
• System cost (excluding man-power for program conversion) 12-16MF (1980) (6.8-9M€ 2017 value)

IBM A virtual memory system, 32-bit VM-CMS system was proposed. A major weakness was the necessity to pre-allocate disk space to users, rather than the more common and flexible pooled space of the other systems. Networking to the existing ILL systems required major development to adapt to the IBM proprietary SNA system.. At that time there were not many systems in use elsewhere. With two American companies this proposal was eliminated at an early stage.
ICL An ICL 2980 32-bit system was proposed by ICL running VME/B system. This was used in several UK universities, with varying degrees of success (there was considerable discontent at Oxford). With a commercial background the system mostly used intelligent data entry terminals (non-graphical) which performed local editing etc, rather than simple ASCII terminals. To satisfy the data disk configuration required (2 GB) required 16 disk units. The new computer room was dimensioned sufficiently to contain such a large system. The sales group had its own technical group, effectively isolating the development engineers from any contact with clients.	ICL 2970 peripheral units ©2009 Roy Pedlar
CII-Honeywell-Bull A Multics system running on a bi-processor DPS2 36-bit computer, similar to that at the university of Grenoble was proposed. The system was completely virtual, mapped between memory and disks, and was almost completely written in the high- level language pl1. There had been problems at the university without a guaranteed power supply, and the inevitable electrical storms in summer. This was resolved by copying systematically modified pages, or segments, to tape. Rebuilding the system by system engineers still required several hours. The floating binary format was distinct from the 36-bit PDP10, without the 128 exponent bias; there were hence two representations for 0. As in Unix systems there was a similar hierarchical file structure, and and command given was interpreted as a valid pl1 instruction to seek out and start execution of a segment so named.	Prototype of a Multics Honeywell-Bull 6180 (1973) ©<multicians.org@gt;
Siemens The Siemens BS2000 32-bit (IBM machine code) system was proposed by this German company. When a number of practical problems were identified by the ILL, Siemens staff were very willing to introduce changes to simplify introduction and use.	Siemens BS2000 in 1980 ©CEA, Centro de Estudios Administrativos, Salamanca
DEC Initially a bi-processor KL1099 was proposed, with similar software to that already in use at the ILL,, offering a simple transition from the existing system.

As can be seen, these mainframe where huge sets of cabinets connected through a web of cables. Thus a large air conditionned room was required and this was anticiped in the new building which was being built.

The sales groups from each manufacturer visited the ILL repeatedly;

• ICL used the ILL to train its sales staff with unnecessary visits, and hosted a visit to Luxembourg to the EU datacenter to try out their system (and intelligent terminals) with sample ILL programs.
• We were given access to a Multics system in the US using the early ARPAnet, using a dial-up 30cps modem. We moved several programs using the Texas instruments silent 700 terminal with cassettes for file transfer.

A set of programs were made available to the manufacturers who were requested to compile and run them with a log file of terminal response times for each operation for benchmarking. These files were then examined at the ILL; there were clear difference in system response times for simple environment assignments to the time taken to execute compute tasks within programs. Some long calculations were included to assess innate computing power.

There was considerable pressure from the British to turn to ICL; similarly the French emphasised the complementarity with CICG in Grenoble.

To quantify the cost in manpower for changing systems it was decided to examine twenty programs from the regular experimental activities. Each manufacturer was provided with tapes of source, and solutions, which included Tektronix 4010 graphics. They were invited to ask for help at any stage. The transformation exercise was set to last some twelve weeks, at the end of which they were invited to demonstrate their programs at the ILL.
ICL allocated two average programmers, of a quality which might be expected to aid a transition over a period of up to a year. CII introduced a young American, to work on the CICG system. We never met the whole team which Siemens allocated to the project.

 At the end of the trial period representatives all met up at the ILL. Results were recorded on a blackboard. Three programs were accepted as being untransferable since they used DECtapes, which would have to be phased-out.

• ICL brought a 2400 baud modem; one programmer ran the programs successive at the ILL, while his colleague in the UK managed any changes.
• The Siemens team (lead by a daunting Frau Rechner!) brought several terminals and 2400 baud modems and ran their tests in Munich, with a spare "hot" computer available in case of problems. At 16:00 Frau Rechner (Siemens) insisted the room with the blackboard be closed!
• Unfortunately the Multics at the University had failed in the morning; there was much reduced time for the young Whizz-kid to show his work.

In practice the initial results showed all machines were technically acceptable, if program transformation could be funded. The usability of the machines could be judged by comparing the scripts for running programs which were actually proposed in the tests. As an example, Siemens and ICL could not explicitly open files within Fortran programs. Siemens were prepared to add this feature. ICL introduced a pause into the program, where the user could enter the required association of the filename and I/O stream, then continue execution. To facilitate use ICL effectively used command files to hide the system complexities, but scripting these also had to be costed. Over the long period of program tests ICL identified that a considerable fraction of total power was expended in terminal handling for their intelligent terminals. By eliminating this code to treat the simpler ILL alphanumeric and graphics terminals this was reduced by 60%, and enabled them to offer a cheaper ICL2972 model, rather than the 2980.

At the autumn steering committee the Directors actually had available one proposal for CII and one for DEC, but neither was actually presented at the meeting.
DEC then announced a new model, the KL1091S, and new disk sub-system, which would be manufactured in the UK (at Ayr). The newer configuration was priced significantly lower than previously, partly due to the inclusion of two RP20 disks, each with 1GB storage on two fixed spindles.

Finally a decision was made to acquire the updated DEC system.

A minimum configuration (cpu, RP06 600MB disk, 2 tape-drives) was delivered shortly afterwards, and housed initially in the ex-NICOLE computer room while the new computer building was being completed. During this period of several months when both machines were operational the final installation was prepared. The 800bpi tape library was rewritten using the PDP11/55 concentrator which had two dual density 800/1600bpi tapes.

Final configuration

• DEC 1091S 1M word memory, 3.4 x speed of KI 1070, PDP11/40 front-end 2RX01 8" floppy disks (boot)
• 2 RP20 winchester disks (2x1GB), 2 RP06 disks (2x180MB) 4 self-loading tape units (1600/6250 PE tapes)
• 2 DN20 communication processors (actually PDP11/34) 2 LP20 1200 lpm printers
• Benson Graph pen-plotter
• LA120 printing console
• Slow peripherals from the KI10 were reused, including 2 LP10 printers.
• Final cost about 8MF (1980) ( ~4.5 M€ 2017)
• Approximately 16 VT100/PT100 alphanumeric terminals were acquired for users 4 Pericom GRAFPack raster graphics terminals
• The MICOM/SATELCOM terminal switching system was installed and operational by the end; this started the tendency to cable up the scientists offices for terminal access, and losing the sharing aspects imposed by the two rooms of terminals.

Back to main page       Top