Jump Over Left Menu
Atlas Computer Laboratory
Some Aspects of Current Operation and Reseach
Atlas Computer Laboratory
WE have put together this collection of papers with the aim of showing the range of activities that go on in a big and busy computing centre. As members of the Atlas Laboratory we have to meet service obligations - stated simply, to get as much work as possible out of our machine - and to tackle problems of compiler writing and general software construction on the solution of which the quality of the service depends. We have also a number of individual research projects which depend upon the availability of a large fast computer.
The papers have been kept short and, we hope, self-contained; and we would like to think that they will be found interesting by a wide range of readers. Any of the authors will, of course, be happy to give more information about his work to a reader who would like to know more about it.
Jack Howlett Director
Atlas Computer Laboratory Chilton, Didcot, Berkshire 14th September 1966
The Atlas Computer Laboratory
THE Atlas Computer Laboratory was set up by the British Government in 1961. It was originally administered by the former National Institute for Research in Nuclear Science and came under the Science Research Council in April 1965, when that body was created as part of the Government's implementation of the proposals of the Trend Report on the organisation of civil science. The aim of the Laboratory is to provide computing facilities on a large scale - with all necessary supporting services - to research workers in universities and government laboratories; more specifically, to be a place to which they could turn when faced with problems needing more computing power than their local installations could provide. A cardinal feature is that no charge is made to university users for any work done for them. Government users are charged at a rate which represents the cost of operating the Laboratory.
The Laboratory is in Chilton in Berkshire on a site adjacent to the Rutherford High Energy Laboratory of the Science Research Council and to the Atomic Energy Research Establishment (Harwell) of the United Kingdom Atomic Energy Authority. Its main equipment is a large Atlas installation which was ordered from Ferranti (later merged with 1. C. T.) in the summer of 1961. A building was designed with the special needs of the computing service in mind and was ready for occupation in January 1964. The machine was installed during May and June of that year and a regular one-shift service was started in October. This has been extended as demand has increased and at the time of writing we are running three shifts five days a week; we expect to take up the weekend gradually over the next twelve months.
The installation is made up as follows:
- Main Frame
48K core store (2µs cycle time)
96K magnetic drum store
8K fixed store (0.4 µs access time)
1K working store
The word length is 48 bits. The fixed store holds the basic routines and uses the working store as working space.
- Magnetic tape
16 Ampex TM.2 decks
2 IBM 729 Mark IV decks
Card readers: 2 ICT 600 card/minute
Paper tape: 2 Ferranti 300 character/second
Printers: 2 Anelex 1000 lines/minute, 120 character /line
Card Punch: 1 ICT 100 card/minute
Paper tape: 2 Teletype 110 character/second
This is backed up by standard equipment for tape and card punching, card reproducers, interpreters and sorters. There is also an off-line Benson-Lehner Model J graph plotter driven by magnetic tape, with a library of programs which make it easy for a user to get output in the form of point plots, histograms or continuous curves, including stereoscopic pairs for display of three-dimensional structures.
The machine runs at an average speed of about 350,000 instructions per second. Floating point addition takes 1.8 to 2 µs, multiplication 5.9 µs, organisational instructions 1.6 to 2 µs. In machine code, inversion of a 100 × 100 matrix takes 17 seconds, sorting 5000 floating-point numbers into numerical order takes 1 second. The operation of the machine is very automatic and is normally under the control of a permanent resident program called the Supervisor, which monitors all its activities, organises the time-sharing of input, output and computation and of a number of programs held in the store at the same time. One part of its action deals with automatic transfers between the core store and the magnetic drums, and enables the programmer to work as though the two formed a continuous directly-addressable store of 144K words. It also makes possible the use of many different compilers, and in fact Atlas users can (and do) write in Fortran, Algol, LISP, IPL-V, the Atlas Autocode produced by Manchester University, the Autocode produced originally for the Ferranti Mercury computer and since greatly extended, and several others.
The level of activity can be gauged from a few operating statistics. In a typical week we run 2,500 jobs, input 800,000 cards and 30 miles of paper tape, print 1.8 million lines of output, punch 50,000 cards, handle 1,200 reels of magnetic tape. We have 250 projects on our books from university users and are usually doing work on 100 of these. A recent survey showed that the distribution of the work load amongst the main languages is:
|Compiler||% of Jobs||% of Computing Time|
|ABL (Machine Code)||6||5|
and of job applications amonsts subjects is:
|Area||% of Jobs|
|Biology and Medicine||5|
The Laboratory has been planned on the assumption that the total staff will rise to a little over 100, to be reached during 1967 when the machine is running seven days a week. This does not include the computer maintenance engineers, who are I. C. T. staff working under a maintenance contract.
There are three technical groups:
- The Operations Group receives work (some of which is in manuscript and needs to be punched on to cards or paper tape), processes it through the main computer and any auxiliary machines, such as the graph plotter, and dispatches the results to the users. Any running difficulties which arise are dealt with on the spot by a liaison section in order to avoid delay wherever possible.
- The Programming Group is concerned mainly with basic software for example, changes or extensions to the Supervisor to incorporate new ideas or extensions to the installation - compilers and library programs. It is building up an extensive library of programs in Fortran and Algol. In addition, it gives help and advice to users and is responsible for educational activities such as programming courses and lectures on relevant topics in mathematics and computer science. A number of programmers are also engaged in more specialised activities such as developing a system suited to the analysis of statistical experiments and an information retrieval project.
There are posts for individual research: the Laboratory has given fixed-term contracts of employment to research workers of established reputation whose particular interests demand the use of a large-scale computer or who are interested in exploiting the powers of the machine in novel ways. It has been the policy wherever possible to arrange that the holders of these posts also have some academic connection, such as a University or College Fellowship.
There is, of course, a small group to look after the administrative needs of the Laboratory.
As can be imagined, we have had to give a lot of attention to the office-management aspect of the operation of the service, so as to ensure that the very large volume of work (with all its associated paper) is handled quickly, efficiently and correctly. We make as much use as possible of modern office machinery and, of course, we use the computer itself to produce all our accounts and statistics.
A point worth noting is that we expected to have significant numbers of people wanting to spend some time in the Laboratory, to develop large programs or to see large-scale production projects through. We planned the building so as to provide pleasant and practical accommodation for visiting users, in small single offices which can be booked a week or so ahead. These have proved very popular indeed and have made life far easier, not only for the visitors, but also for the permanent staff of the Laboratory, who are thus protected against invasion. With the accommodation problem solved, the presence of many visitors, with a great variety of interests, is extremely stimulating and contributes a great deal of intellectual liveliness to the Laboratory.
Extensions to the Installation
Our experience over the past year has shown that the Atlas central processor, with the Supervisor which is an integral part of the system, is an exceedingly powerful and flexible device which deals smoothly and efficiently with a heavy load of very varied work. To exploit this to the full we want to extend the installation and at the moment we are making plans to add the following:
- A disc file, capacity 16M words, expandable to 30M. This will replace the four tape decks now given up to the operating system Supervisor and compilers, input / output buffers, dump; will hold library programs; will allow programs under development to be kept in the system, thus removing much of the need for punching binary card decks; will provide backing-store for big programs; and will form the storage basis of the on-line system we want to install.
- A small on-line system, initially with 16 consoles, communicating with Atlas through a small satellite computer and the disc. The satellite will do most of the organisation and so leave the Atlas processor free for production work.
- Visual display and microfilm reader, to allow automatic high-speed production of plots, graphs, cine-films; if possible, as on-line equipment controlled by the satellite.
We hope to have all these in operation by about the end of 1967.
We are about to start on an extension of the building, to provide more rooms for visitors, a library and reading room and a new lecture room.
The Laboratory has been providing a computing service for two years, and very intensively for the past year. The use made of this service shows how great is the need: we get work from every university in Great Britain, the demand is increasing and there is a steady increase in the general levels of size and complexity of the problems put on the machine. It is particularly pleasing to observe the rising demand from workers in the non-physical sciences - sociologists, psychologists, educationalists, biologists and others - as they begin to appreciate the help they can get from a powerful computer and a quantitative approach to their problems.
The Atlas Laboratory is operating as a national facility for universities, and to a smaller extent for government research laboratories; it is a great deal bigger than anything else to which the universities as a whole have access and is comparable with anything in Europe. The scale of operations is great enough to justify the provision of many technical and administrative services and we have come to the conclusion that these are valued almost as highly by our customers as is the computing power of the installation. Simple things like the supply of cards, paper tape, coding forms and stationery generally are very important; people need ready access to card and tape preparation equipment and to desk calculators for odd checks and minor pieces of arithmetic; help in arranging transport and booking accommodation is very welcome; and of course somewhere to work is essential. On the technical side, and taking for granted the vital need for first-class basic software and library programs, we have found a great demand for a general advisory service for users. We now have a group of three people - shortly to be increased to four - engaged in work which includes correcting minor faults in distant users' programs (thus saving time which would otherwise be spent in posting the work backwards and forwards), giving advice on the use of standard programs and monitoring extended production projects.
Finally, all experience has confirmed the view with which the Laboratory started out, that the presence of research activities in the building and easy contacts with the academic world and with other computing centres are essential to the intellectual health of the members of the Laboratory. A large-scale computing service is really a very sophisticated and highly professional undertaking which makes demands at all levels and is always needing both stimulation and criticism; without these it is fatally easy for the people who are providing the service to become stale and unenterprising, and the standard of the service to decline.
The papers in this volume are placed around the site in the most appropriate place. Links only are provided here.
- Fortran on Atlas: the Hartran System. E.B. Fossey and Barbara Stokoe.
- The Algol System on Atlas. F.R.A. Hopgood and A. G. Bell
- Simulation on Atlas. P. E. Bryant
- The Organisation of Store Utilization for a List Processing System in a Time-sharing Environment. D. B. Russell
- Time-sharing and Atlas. M. H. J. Baylis
- The Analysis of Sample Surveys. J. E. Hailstone
- ASCOP - A Statistical Computing Procedure. B. E. Cooper
- Animal Feeding trials Program. T. N. Gover
- The Atlas Laboratory Information Retrieval Project. R. F. Churchhouse
- The Use of Atlas in the Discovery of a Theorem in the Theory of Numbers. R. F. Churchhouse
- Computing Partitions and Thirteen. A. O. L. Atkin
- Radiative Transfer in Planetary Atmosphere. I. P. Grant and Valerie M. Burke
- Solution of the Relativistic Hartree-Fock Equations. I. P. Grant and Valerie M. Burke
- Crystallography. J. C. Baldwin