THE Atlas Computer Laboratory of the Science Research Council is embarking on an experiment to test the usefulness, of multiple access computer systems. According to the Science Research Council, the project will cost £95,000. By the beginning of 1968 it is planned that there shall be twenty teletype consoles linked to the Atlas computer through an intermediate satellite machine built by the General Electric Co., Ltd. (U.K.). In the planning of the experiment, careful attention is being given to the quality of the system, and the planners are particularly anxious to avoid peripheral troubles of the kind which have occasionally afflicted other multiple access systems - inadequacy of teleprinter links, for example. The satellite computer is intended to take over much of the housekeeping work associated with multiple access systems; experience with Project MAC at the Massachusetts Institute of Technology is said to have shown that roughly a third of the time spent by users of multiple access systems goes on comparatively humdrum activities such as editing and retrieving information. Obviously it is not sensible to bother the big machine with chores like these. In practice the multiple access experiment will have to compete with the other users of the Atlas computer who at present keep that machine more than saturated. (For every job taken on, two are turned away.) The laboratory is co-operating with International Computers and Tabulators, Ltd., in a study of the habits of computer users when multiple access is available. Among the first users of the multiple access facility will be a group at the Atlas laboratory working on artificial intelligence, the group concerned with the statistical planning system known as ASCOP, and a scheme to use-multiple access for information retrieval within the Culham Laboratory.
The laboratory itself seems now to have settled down in the building at Culham, Berkshire, which it has occupied since the middle of 1964. At present the computer is running three shifts on five days a week, and the weekends are likely to be used as well in the course of this year. The computer itself is now in service for between 95 and 98 per cent of the available time, and half of the unplanned shutdowns seem to be due to failures in the 48K core store with which the machine is provided. The experience of the past few months has shown that, in a typical week, something like 2,500 jobs may be undertaken at the laboratory. The principal customers are universities, and there are 250 university projects on the book, most of them in mathematics, the physical sciences and engineering. The laboratory also works for some industrial users and for government research laboratories. It has, for example, been concerned with numerical forecasting for the Meteorological Office.
Computers may be eventually used for predicting local rainfall after research by F. H. Bushby and Margaret Timpson of the Meteorological Office, Bracknel1. This was discussed yesterday at a meeting in London of the Royal Meteorological Society.
The unit of prediction might be a square with sides of about 25 miles length. Good agreement has been obtained in trial calculations between predicted and actual rainfall for amount and location.
The limitation is that it takes eight hours' work on the Science Research Council's big Atlas computer to produce a forecast 24 hours ahead. To turn research into regular practice would require a computer 30 times faster than Atlas but it is regarded as a landmark in numerical forecasting and is believed to place Britain in the forefront of research in this field.
BRITISH meteorologists are delighted at the success with which Mr. F. H. Bushby and Miss Margaret Timpson of the Meteorological Office at Bracknell have been able to forecast rainfall by computer. The essence of what they have been doing was embodied in a paper read to the Royal Meteorological Society on February 15 and soon to be printed in the society's Quarterly Journal. Briefly, they have used a three-dimensional model of the atmosphere in which the possibility of precipitation is adequately taken account of. Numerical integration of the equations representing the model atmosphere has been carried out over a three-dimensional grid of points, with those at ground level separated by 40 km. Thus the physical scale of the weather forecasts obtained is comparable with that of the frontal systems with which precipitation is usually associated. Most of the numerical forecasting systems already in use depend on an integration network whose dimensions are more nearly comparable with the scale of large-scale phenomena such as anticyclones and the like. This system of equations has been used to produce a number of 24 hour forecasts for the British Isles and north-west Europe. A full integration takes 8 hours on the Atlas computer at the Chilton Computer Laboratory of the Science Research' Council, which means that the meteorologists will now join the queue of those who would like to see the early coming of still faster computers.
The success of the few 24 hour forecasts so far produced has been impressive. In the prediction of the movement of a small wave depression across the south of Britain on December 1, 1961, for example, the progressive deepening of the depression was forecast, although the numerical value of the lowest pressure was 3 mbar too high after 12 hours and 3 mbar too low after 24 hours. At least a part of these discrepancies seems to be accounted for by large scale oscillations in the atmosphere with a time period of several hours, but fortunately their amplitude does not increase with time. One of the particularly cheerful features of the work is that an initially smooth network of observations can yield a sharp frontal system with the passage of time. Forecasts of actual rainfall agreed well with reality, although the axis of the predicted belt of rainfall was farther north than it should have been. Because the model does not, as yet, include the effects of convection, some thunderstorms were missed. It is acknowledged that the consequences of friction and topography are not included in the atmospheric model, which means that rainfall may not be fully accounted for in the work so far carried out. None of this, however, has prevented meteorologists long.trained to keep a quizzical eye on proposals for numerical forecasting from saying that the Bushby and Timpson article is a milestone in numerical forecasting.
