Mike Baylis was a member of staff during the early years of Atlas with a particular interest on the hardware side. He was involved in the assessment for the Sigma 2 system that provided interactive access to Atlas and also designed the Engineer's Assistant that provided semi-intelligent support to the Atlas Engineers.
Mike was previously at Ferranti's and was a strong advocate for a top-end replacement for Atlas and strongly supported the purchase of a CDC 7600 later in 1967 and later the CDC STAR-100 bid. The purchase of the ICL 1906A he felt was the wrong machine for Atlas. He left to join ICL and later CDC. He continued to live in Sparsholt until his death at a relatively early age in 1987. He had a son Alex (named after Alex Bell) and a daughter Lucy.
The letter below written on the 13 January 1967 to Jack Howlett indicates the strength of his feeling on the Atlas replacement. Many of the Atlas staff had similar views but probably did not express them so elegantly. In June 1967, Bob Hopgood visited CDC at Minneapolis and Chippewa Falls to get an In Confidence briefing on the CDC 7600 that was reported back to Jack Howlett, Lord Halsbury, Christopher Joliffe and Brian Flowers.
It has become well known that I am vigorously opposed to the Atlas Computer Laboratory buying a 1908 computer. Anyone who holds strong and definite opinions - on anything - is in some danger of being considered irrational or bigoted, and then discredited and discounted. I have written here a collection of facts and opinions, relatively rational, which I know and believe, about the Atlas Laboratory, and the present and future state of the British computer industry. I don't particularly care about being discredited but I intensely dislike irrational decisions and I do care about the above three topics.
There are three major companies left in the digital computer market, Elliotts, EELM and ICT.
Elliotts: are dedicated to the small and medium sized computer market. Their current range, the 4100 series, has been described in my paper Satellite Survey. They have about 30 orders and have delivered around 20. None of the fastest machines (the 4130) have been delivered. The design of the machines is fairly restrictive - not quite as restrictive as the ICT 1900 range - but it could certainly cope with at least one higher member of the range. Elliotts have opened an integrated circuit factory in Scotland and presumably the fruits of this in the next few years will be the 4140, and perhaps a 4150. With suitably fast stores these should be competitive machines and extend the life of the range into the early 1970's.
English-Electric-Leo-Marconi: The EELM System-4 is a range of machines which are IBM 360 compatible. Up to the 4-50, the design and technology are provided by the American company RCA. The technology is multi-level integrated circuits. RCA market the same range in the USA but with conventional components under the name of the Spectra-70 range. EELM are solely responsible for the 4-70 and 4-75. Although these are copies of the IBM 360/65 and 360/67, EELM deserve some credit for at least trying to exploit the know-how they have bought. It has been said that to copy the fashion is the only way to be certain of never being in fashion. However, RCA's objectives are to produce machines identical to IBM's, to undercut the price and perhaps to increase the speed of operation. As far as can be seen, the 4-75 is slightly slower than the 360/65 but it is a lot cheaper, up to 30% I believe. University College's 360/65 cost £420,000; an identical 4-75 was quoted at £340,000 but delivery not for one year. (Of course IBM can give educational discount to counteract this for some customers.) I personally can feel little sympathy for a company dedicated to copying. I don't have to make a profit though.
ICT: pay RCA a cool £500,000 per year for the benefits of RCA's research, but they have the sense not to use much of it.
The current ICT machines are the 1900 range, which at present extend from the minute 1901 to the 1906 and 1907, the latter probably being about ½ to 2/3 of an Atlas for one given program. For throughput it is not likely to be more than ½ of an Atlas. At present the first 1907 is four months late on delivery because it is not as fast as specified. The machines are based on the FP6000 a computer intended as a Pegasus successor and designed jointly by Ferranti Packard in Canada and Ferranti's. The FP6000 corresponds roughly to the 1905. Incidentally the odd numbered machines have hardware floating point arithmetic the even ones do not. The 1905 seems to be about 1/5 an Atlas for a given job. Up to and including the 1905 the design is reasonable. It is essentially a small machine design, with 24 bit words, 32K directly addressable, one optional floating point accumulator, and 8 index registers, the first three of which can be used modify the address field. There are about 64 basic functions. The sales position (6.1.67) is as follows:
| Machine | 1901 | 1902 | 1903 | 1904 | 1905 | 1909 | 1906 | 1907 |
|---|---|---|---|---|---|---|---|---|
| Orders | 254 | 171 | 89 | 60 | 30 | 16 | 4 | 10 |
| Delivered | 22 | 95 | 55 | 30 | 19 | 13 | 0 | 0 |
| Home | 171 | 97 | 59 | 43 | 21 | 10 | 4 | 7 |
| Abroad | 83 | 74 | 30 | 17 | 9 | 6 | 0 | 3 |
The 1909 is basically the 1905 without the standard interface (I think) so if we regroup the sales into sets, we have:
| 1901 | 1902/3 | 1904/5/9 | 1906/7 |
|---|---|---|---|
| 254 | 260 | 106 | 14 |
It will be noticed that sales of the 1906/7 have been very low. In fact of the 14 orders, two are by ICT themselves and three have been placed by the Flowers Report. Is this because there is no market in Britain for a machine of half-Atlas power or is it because it is not a good buy? Well, ICT admit that the EELM 4-90 is a 60% better buy on a performance/cost basis, and I think that EELM have more than 14 orders for the 4-70.
ICT intend to produce integrated circuit versions of the 1900 range over the next few years, of approximate powers 1, 2½ and 4 times the 1906 power. This is clearly a sensible thing to do and should ensure that 1900's continue to sell up to around 1971. The technology for this will be some modified form of their own 50 Mc/s modules made by Ferranti's or Motorola MECL modules from the USA. The latter are cheaper but slower; their main advantage is that they are readily available. There is some evidence that 150Mc/s circuits will be available in the next two to three years which would allow say a 10 times 1906 machine to be built for (say) 1972. It is my opinion that the design cannot stand up to the requirements for a large machine. Given any technology, a computer could be designed to be between two and four times the speed of the 1900 design.
To counteract this ICT propose a 1908 machine which will consist of up to four processors sharing a common core store. The proposal is basically to extend the 1900 range life a little longer. The difficulties and failures of other multi-processor manufacturers are ignored. These include DEC who failed to get two processor PDP-6 working, and currently GE who are doing a 1700 wire modification to their two processor 645 machine at MIT. GE have also withdrawn the machines from sale for the time being.
At this moment IBM offer the 360/65 at a comparable price to the 1907. They have the 360/75, at double the 65 power. Next year they will be delivering 360/91's (possibly) reputedly about 30 times the 1907 power. CDC offer the 6400 at not much more money than the 1907 although it is around 3 to 4 times as powerful. They have the 6600 at 6 to 8 times the 1907 and expect to have the 6800 in 1969 which will be 24 to 32 times the 1907. Other US companies such as UNIVAC, SDS and RCA will be competing in this market. Note that the CDC 6600, with conventional components is 6 to 8 times as powerful as the 1907 and this machine has been running well for a few years - there are around 20 delivered. Also note that IBM and CDC are not building multi processors yet.
The European market is about as big potentially as the current USA market. This would suggest that over the next few years between 50 and 100 machines of about 6600 power could be sold in Europe. (CDC have over 60 orders for the 6000 series, IBM have over 10 for the 360/91). In England there is the Atlas Laboratory, Aldermaston, Harwell, two regional Centres - presumably at least three more future regional centres - the Meteorological Office, a handful of large industrial concerns and defence establishments that one can think of offhand as potential large machine customers.
The advantages of large machine computing are mainly in the low performer cost but also in the range of jobs that can be coped with. One hour of 7090 time, say twenty minutes of Atlas time, works out at £18 on the CDC 6600. Atlas can take jobs from the minimum to those using 120K of store, 16 magnetic tapes and lasting many hours. British Petroleum claim to have saved themselves £M1 on the London Atlas last year. The disadvantages are the high design and production expenses, both on hardware and software. ICT think that by extending the 1900 range they will save money on this. However the 1908 will never be a competitive machine and could only be sold by the restrictive practise of the Ministry of Technology forcing it on a captive audience. Further, what do ICT do in 1972 when they have finished this uncompetitive machine, and the general 1900 range is not selling? Presumably they will then, far too late, start designing another product range. I have argued that Atlas would be a good basis on which to build a range of machines starting at 1 to 2 × Atlas and going up to 10 to 20 × Atlas. Large parts of the software could be preserved and a lot of the system design. In parallel, a new commercial range, perhaps based on the Illife Code-word machine, could be in development to supersede the 1900's. The advantages of two ranges, slightly overlapping, is that at least one is in production all the time. This is the sort of thing that the Ministry of Technology should be encouraging, rather than supporting a second rate time delaying project which many people in ICT do not even want to do.
Alternatively it should be admitted that the British computer industry will never be competitive with the USA in the large machine area and permission given for American machines to be bought where a good case can be made. I think this would be wrong because, although it is getting harder to believe, Britain is still ahead in many aspects of large machine design.
The question is simply whether the laboratory has a future or a great future. In my opinion if it buys a 1908 it has the former and I for on~ will not be terribly interested in it. The Laboratory consists of a first rate Operations Group and a reasonably high powered Programming Group. The Operations Group are responsible for providing a University research computing service. The Programming Group, besides backing this up with systems programs, compiler writing etc. do enough research in computer-oriented science to warrant a great deal of respect. Neither the Operations nor Programming Groups want a 1908 computer, given a free choice, and this statement cannot be said too often at the moment. For various political reasons the Director of the Laboratory has felt obliged to put in an application to buy a 1908. The arguments given to the laboratory afterwards are that there is nothing else on the British market and the Government will not let us buy an American machine. These arguments do not seem to me to be very strong. I would have preferred the laboratory to lead the way by stating what sort of computer it would like and sticking out for it. After all it has reason on its side, and the interests of its users and the computer industry to consider. The danger of requesting the 1908 is that the Ministry of Technology will assume we really want such a machine. The danger of not saying what we would really like is that we give no clues or guidance to the industry to aid its future developments. The danger of not providing the best computing facilities for University research is that research projects suffer. (After all, a University research project led to the birth of the computer industry in Britain.)
There is no doubt that the Laboratory needs more computing power; I guess it could do with 4 × Atlas power as soon as possible. We particularly need a machine to cope with the very large and long running jobs which Atlas cannot do. To my mind, a very strong case can be put for buying or renting a CDC 6600. In parallel, we could co-operate with ICT on an Atlas like range for the 1970's and place an order for the largest machine. If ICT later decide not to compete in the large computer market then the Laboratory could get a CDC 6800. Either way, University research and our own work would have the best environment obtainable in the next 10 years.
Finally, I would like to see the Laboratory co-operating with and preparing to supplement the regional centres. For example, we could provide some computing backup for the multi-access users at the Edinburgh centre, if we had a powerful enough and flexible enough machine.
M. H.J. BAYLIS
13.1.67
