Formation of Mathematical Services Department
The Royal Aircraft Establishment began to develop an important computing centre during the late 1940s. Hollerith machines were installed in 1944 as the basis for a Computing Laboratory which was then part of the Structures and Mechanical Engineering Department. R.A.Fairthorne led the work of the laboratory. A Mathematical Services Department (MSD) was formed in 1948. The first head of the new department was Dr Stuart Hollingdale who had joined the Aerodynamics Department of RAE in 1936. The total strength was 5 and the only equipment available was the Hollerith punched card machinery mentioned above. MSD went on to become a well-known computing centre in the late 1940s and early 1950s.
Dr Hollingdale published a review of the work of his department in April 1952. This surveys the main activities of the department in the period January 1951 to March 1952. The review gives the terms of reference of the Mathematical Services Department. These were:
At that time, the Department had three sections: general computing and punched card equipment, automatic digital computers and devices and analogue computers and devices.
Concerning general computing there was stress on improving computing efficiency through use of correct numerical methods, layout of the work, matching of the desk calculating machine to the job and to the operator. The Department had a general responsibility for education and training. Staff gave general lectures and also assisted the then RAE Technical College in the training of new scientific assistants.
Moving on to automatic digital computers, the author of the report recognises their development as the most significant event in the computing world in recent years. Moreover such machines were becoming increasingly reliable. The report goes on to discuss High Speed Electronic Machines as follows:
The 1952 review also discussed the department's involvement in the design and construction of a sequence-controlled calculator which became known as RASCAL (RAE Sequence Calculator) . Sequence-controlled calculators lacked stored programs and generally held their instructions on some form of semi-permanent 'external' storage such as punched paper tape. The RAE machine was decimal and also envisaged a certain amount of external storage, originally based on relays. RASCAL eventually included a drum store and decatron valves. RASCAL was nearing completion in 1956, with the Plessey Company involved in the development under contract.
October 1956 saw a recommendation that the contract with Plessey be terminated although the work had not been completed. The equipment so far produced had been tested and many faults found. In view of the work required to complete, it was recommended that work be stopped and the work programme item be deleted.
Use of ACE Pilot
From the end of 1951 to the early summer of 1953 some work was farmed out to the ACE Pilot Model Computer ( the forerunner of DEUCE) at NPL Teddington. In the summer of 1953 two members of the department learnt to programme this machine and from this time until May 1955 work was carried out on the Pilot model using programmes written by members of the department. In November 1953, Mr E.J. (Ted) York came from the Maths Division, NPL to take charge of the Computing Division at RAE.
In November 1955 the first quarterly report of the department's R&D work appeared. The divisional structure had developed to
DEUCE arrives - in duplicate
The first DEUCE computer was installed at the RAE in May 1955. It ran its first main job on 13 June 1955. The machine was soon working some 13 hours per day. In August 1956 a second machine was ordered. This was delivered in December 1956 and was installed and commissioned by the middle of January 1957. The first machine became known as Gert, the second, as Daisy.
By May 1957, Operational Division (OD) of MSD was reporting that the two machines were available for use. Gert had been overhauled by English Electric Computers to increase serviceability but after many weeks of overhaul the machine had failed the acceptance tests. Pressure of work led to the machine being accepted pending Daisy becoming fully operational. Soon the overhauling of Gert was to recommence until the acceptance test was passed.
By the end of September 1957, the two machines had completed 4,582 hours production work. 86% of this total (3948 hours) was used by principal users as follows:
|User||Hours||Percentage of Total|
|Guided Weapons Department||1744||44.2|
|Aircraft & Weapon Firms||205||5.2|
July 58 saw the announcement that Maths Department was to install a Ferranti Mercury computer in Building R14. The layout had been completed and drawings were in progress. Delivery was scheduled for the first quarter of 1959. Ferranti delivered the computer on 1 April 1959 and commissioning began in May. There were setbacks. Both drums had to be returned to works for repair and several refrigeration faults developed. This led to a delay in commissioning and acceptance tests were scheduled for July/August.
Mercury completed acceptance trials in August and became operational in September. A notable feature in the layout of the RAE Mercury was the arrangement of the control desk and control panel; these were located in a control room which was physically remote from the computer. It is believed to have been the only Mercury installed with this feature - in this respect, well ahead of its time.
During the last quarter of 1959, teething troubles were still experienced especially with the magnetic drum units. New type drums would be available during 1960. By July 1960, Mercury was being used on a regular 1.5 shift basis during week. An intensive period of staff training had been undertaken to provide the additional engineers required.
Mercury takes over from the DEUCEs
By mid 1961 the demand for evening shifts on the DEUCEs slackened as new work moved to Mercury. The reliability of the Mercury was better than in 1960. At the start of 1962, the two DEUCEs were running reasonably satisfactory. Mercury reliability was at a high level. 96% useful time was recorded - useful time being defined as total time less scheduled maintenance.
In July 1962 it was reported that the introduction of a shift system had led to increased serviceability on the two DEUCEs and increased reliability. Mercury maintained good reliability. Two extra magnetic drums (making a total of four) for the Mercury were delivered in June and were soon commissioned and in use. However, increased valve failure rate on Mercury had resulted in lower serviceability figures (94%) by January 1963.
The first half of 1963 saw a drop in serviceability on Mercury because of increased thermionic diode failures. Valves were mounted in such a way that it was almost impossible to extract them without breaking. It was decided to change banks of diodes as part of routine maintenance.
In summer 1964 a modest increase in Mercury utilisation was observed with a corresponding decrease in the use of the DEUCEs. The load on the DEUCEs continued to decrease throughout the rest of 1964. However, during 1965 a combination of extra DEUCE work and poor serviceability resulted in No 1 DEUCE being switched on again but it would be shut down again in the New Year (1966). Disappointing serviceability of Mercury (84%) was recorded in January 1966.
The DEUCE service closes down
By July 1966 demands for DEUCE time had decreased to 27 hours per week. It was no longer economic to run. There was a plan to shut down DEUCE by the end of the year. On Mercury, modifications to increase reliability were being made. Serviceability was 89%. Eventually both DEUCEs were switched off on 30 November 1966. Mercury modifications continued with satisfactory results (88%).
Mathematics Department continued to run the Mercury computer until it was finally decommissioned on 30 October 1970. Its building was required for the installation of a new computer.
Use of extramural machines
During the mid 1960s, the department began to use off-site machines. The July 1963 report observed that some use was being made of the Atlas machine at Manchester University and of the IBM 7090 at the IBM London Computing Bureau. This use continued into 1964. By June 1964, the RAE use of the Manchester University Atlas had grown. A continual postal service was in operation and the Atlas time used was equivalent to 1,000 Mercury hours.
Work on Atlas continued to increase steadily. In January 1965 it was reported that more than half of the RAE computing service work handled by the Maths Dept was being done on this machine.
A year later Atlas computers at Chilton, London and Manchester were all being used. Two thirds of the RAE's computing was being done at these sites. In the first half of 1966 an on-line data link to the London Atlas was installed (mainly for short but urgent jobs). The on-line link to London worked well. The DEUCEs closed at end of November 1966. 75% of RAE's work on the Atlas machines by the end of the year. By the middle of 1967 this figure had risen to 80%.
The ICL 1907
In January 1964, the formulation of requirements for the new general service for RAE and plans for the assessment of tenders were reported as in progress. A new building alongside R14 would house the new machine. RAE based its case for a new machine on the combined Mercury and Atlas workloads, expected rates of growth and the projected computerisation of some of the RAE Secretariat's work - for example, salaries, wages, cost and stores accounting. Two separate machines were originally envisaged, one scientific and the other Secretariat. Ultimately these were united into one requirement.
Throughout 1964 and 1965 the progress of the requirement over the various bureaucratic hurdles was slow. However, by July 1966, RAE had ordered the ICL 1907 computer and expected delivery in May 1967. Eventually ICL delivered the new machine in Autumn 1967 and it was commissioned on 1 November 1967.
[Strictly the machine delivered was an ICT 1907. ICL did not come into existence until 1968 when International Computers and Tabulators Ltd (ICT) and English Electric Computers. merged into one company. ]
Early aims were to take over the Atlas work as quickly as possible and to start the Secretariat work described above. During November and December that year, the department ran the machine on a single shift basis. Meanwhile the organisation structure and management techniques were being worked out in detail. By the turn of the year the department was reporting that there had been some teething troubles but the machine was running reasonably well. Delivery of the magnetic disc storage was still awaited.
The service proper began on 1 January 1968, also on a single shift basis; second shift working started on 1 December 1968. It is believed that the RAE 1907 was one of the first Civil Service machines to run on a formal shift basis.
As well as operating the computers, the department played a significant role in writing application software. In February 1956 Computing Division (CD) had the following tasks:
A report in May 1957 noted CD studying the application of linear programming techniques to organisational problems in the aircraft industry.
Computing Division was later renamed Maths & Computing which was split into two divisions in 1961. One took the title of Mathematical ( & Statistical) Services Division and was responsible for numerical analysis, statistics, programming and the running of the DEUCE and Mercury computers. The other division became known as Applied Mathematics and concerned itself with the larger and more mathematical problems of engineering mathematics. The two divisions were later to merge in 1972.
The work covered mathematical techniques, particularly those techniques of engineering mathematics involving the use of computers. These techniques were studied by working on actual problems posed by other departments of the RAE rather than by the study of 'research' problems, The work was mainly in the following six areas
The study of Numerical Analysis as such was never a major concern. Techniques were studied or programs written as requirements came up. For basic techniques or programs such as matrix programs the divisions relied on the work of others such as the Universities or the Numerical Analysis journals. When work on Numerical Analysis was undertaken, it was usually because the problems were of an engineering nature or because the programs needed were not available elsewhere.
In statistics, consultancy was provided to other departments. Detail of the work is available in reports written in other departments, notably Materials Department and Structures Department. Several large data processing projects were undertaken, one of the most important being the work of E V Hartley on Anthropometric Studies.
The solution of partial differential equations was a major field of study for the applied divisions. Most of the early problems were posed by Aerodynamics Department and were solved in collaboration with members of that department. Most of the problems covered a wide range in classical applied mathematics.
Optimisation work was on specific problems, the most important of which was the preliminary design of a transport aircraft in collaboration with Aerodynamics Department. Additionally optimisation programs were written for other groups in the RAE.
The work on dynamics and control theory was mainly on stability theory and filter theory with some exploratory work on optical control. Some work was done on communication theory and signal processing.
Computing Division undertook the department's traditional training role for the whole establishment. In the very early days of the department, training was given in the use of desk calculating machines and the proper layout of work. When DEUCE arrived, staff were trained how to construct DEUCE programs. Scientific Assistants from customer departments were trained as DEUCE and punched card operators. Training in numerical analysis and computing techniques was also undertaken.
With the coming of Mercury in 1959, CD prepared a Mercury autocode course. By January 1960, seven three-day courses in Mercury Autocode had been given. A considerable number of programs had been prepared. there had been an Increase in demand for help and advice in use of the autocode system.
In July 1960, it was noted that much advice in autocode programs was still required and that a series of short courses in PIG2 had been completed. A fuller course in PIG2 was to start shortly. By New year 1961, two courses in Mercury Autocode and two on PIG2 programming had been completed. Additionally instruction courses in machine operation were being given.
In July 1966 it was reported that, four courses (2 in ALGOL and 2 in FORTRAN) had been arranged for the impending arrival of the ICL 1907. In January 67 training in PLAN for the ICL 1907 and Mercury Autocode was being arranged. The last of a series of 50 Autocode courses was held later that year. FORTRAN courses (of similar scope) were to be started in 1968.
A Programming Systems Division was formed in the first part of 1964 to
In January 1965. it was reported that work was proceeding on CHLF3 Autocode translator for the Elliott 503 computer (this is believed to have been a Structures Department machine) and was expected to complete in August 1965. [CHLF stands for CERN - Harwell - London - Farnborough, the participants in the production of that particular version of Mercury Autocode.] An Autocode to FORTRAN translator was also being prepared. The January 1966 report noted that the CHLF3 compiler (on 503) was completed and in limited use. Work on the Autocode to FORTRAN translator was still in progress. A review of simulation languages was being undertaken.
By summer 1966, consideration was being given to the implementation of a multi-access system. The CHLF3 compiler for running programs on Elliott 503 was reported as being fully tested and in general use. The review of simulation languages had found that CSL, available on the 1907, is one of the best!
By the start of 1967, details of ICL's proposed operating system for the 1907 were being obtained. Experience was being gained with CSL. The format of acceptance trials for the ICL 1907 was being worked out and an EMA (Extended Mercury Autocode) syntax checker for use with remote consoles was being developed.
Summer 1967 saw the development of a multi-access system (MINIJOSS) to enable 5 teletypes to be on-line to Mercury. This was to provide experience for MOP (Multiple On-line Programming).
In January 1968 it was observed that satellite programs were being written in FORTRAN. The operating system for the 1907 was to be that developed by Queen's University, Belfast at first. Then this would be replaced by GEORGE3, G3. Two members of the division had been seconded to ICL - one to work on the EMA disc compiler and one on G3.