The main solution to the immediate shortfall in interactive facilities for engineers was to be the acquisition of a large central computer system for installation at Chilton supporting 60 interactive users and a user population of 600 at a cost of about £1M (see Section 3.2 and Section 7.2 of the Technical Group's Report).
The Tendering exercise for the Interactive Engineering Facility to be installed at Chilton started around December1975. There were quite a few potential suppliers and reading manuals and having meetings with manufacturers continued until February 1976. The work was divided up so that each member of the team had responsibility for one manufacturer and one aspect of all systems. This two-way split was useful in focusing attention on particular aspects.
A major problem was understanding the manufacturer's terminology. A directory on one machine was a catalogue on another and a file on a third. Loaders were sometimes called binders.
The group had quite a lot of expertise in generating benchmarks for the replacement to Atlas and a set of five programs were assembled to form the batch benchmark. Most had been used in the past and and were known to give a good assessment of the FORTRAN system on a machine. They had been stripped down and massaged so that they ran on most machines without unusual behaviour.
The interactive benchmark was a much greater challenge. Most of the machines being considered had the ability to inject a synthetic benchmark into the system either through its front end processor or using a second machine to simulate the terminals attached to the system. The synthetic benchmark was divided into editing/compiling scripts and interactive program scripts. A program was constructed in FORTRAN and data-driven to give any desired combination of processor, interactive and filestore behaviour.
The behaviour specification of the workload was produced based on the requirements in the Technical Group's Report (Tables 7.1 and 7.2). Scripts were then produced, modified, tested and a subset of the benchmark run on the 1906A to check it out.
Discussions were had with DEC, ICL, Univac, CDC, Burroughs and Honeywell. There was some pressure to buy British, if possible, and both an ICL 2900 and the ICL 1906A were offered by ICL. The benchmark was run on a large ICL 2900 system but performed very poorly. The ICL 1906A was clearly under-powered running the GEORGE 4 operating system. It actually performed much better when the machine was run using the MAXIMOP operating system. David Duce ran the benchmark on a ICL 1906A system at the University of Liverpool. However, it still did not meet the criteria.
Existing customers using Univac, CDC, Burroughs and DEC systems were visited both in the UK and USA.
The Univac systems were well tried but had architectural limitations that made timesharing of large jobs difficult. The CDC systems were in a state of flux with major changes taking place in their operating systems none of which were ideally suited for the Chilton system.
After much discussion and attempts to run the benchmark, the only system that seemed to meet the requirements and was potentially capable of running the benchmarks at an acceptable performance was the DEC10 KL system. The benchmark was tried in the UK with an early version of the DEC simulator and it looked as though it would run the benchmark effectively. A formal attempt to run the batch and interactive benchmarks was made at DEC Marlboro on 16 June 1976 by Nancy Ohm of DEC and Bob Hopgood of the Rutherford Laboratory. The SIM11 simulator ran on a PDP11 and kept all the output generate which resulted in a bottle neck when the PDP11 trying to output the results to magnetic tape. A different simulator, SYMLOD, was also tried. Attempts were made to throw output away and output the Tektronix results to real terminals but with no success. It was clear that some of the problems were due to problems in the simulator. A more recent version was being used.
The batch benchmark also failed. A Fortran expression -A*B*C*D was optimised by the compiler to ((A*(-B)*D)*C. When A and B were small, C large and D small, it caused an underflow that did not occur when the arithmetic was done as ((-A*B)*C)*D.
Soon after, SRC finances for the period until April 1977 became a problem and there was a freeze on all capital expenditure until April 1977. Whether funds would be available then was questionable. On the other hand, progress in the area of multi-user minis was greater than expected. In consequence, after much nugatory effort, the decision was made to abandon the installation of a central system at Chilton and concentrate on deploying multi-user systems as soon as possible bringing forward that part of the programme.
