Inside the computer room where the CDC 6600 has served as CERN's main computer. To the left and in the background are magnetic tape Computing at CERN units. In the foreground is the console where the operators are in communication with the A review of the past, present and future of computer's operating system. On the displays computing at CERN concentrating on the large they can see which programs are running central computers which bear the brunt at that time and also the various stage which of CERN's workload. the programs going through the computer have reached.

D. Ball

spared a painful running-in period. The Past The growth in demand for com­ The FORTRAN era came with it. The puting at CERN continued unabated, Computing at CERN started in the problem of the mismatch of me­ amounting to a doubling each year, Autumn of 1958 with the installation of chanical input/output speed to that and this situation, plus the increasing a Ferranti Mercury, which was, for its of electronic computation soon be­ importance of computers in the work time, a modern and powerful Euro­ came apparent and a small ' satellite ' of the Laboratory, necessitated a jump pean-built computer. It was bought to computer, an IBM 1401, was installed in computing capacity preferably of provide data-handling facilities for to relieve the 709 of some of the the order of a factor of ten. The only CERN's physics programme. Its in­ drudgery. In 1962 a flying spot digit­ computer which could give this in­ stallation taught CERN that computers izer to measure bubble chamber film crease was the Control Data Corpo­ are a mixed blessing, requiring a staff was connected to the IBM 709 — the ration 6600 and, in March 1964, CERN of experts to nurse them along parti­ first use of computers on-line at placed its order. The 6600 was cularly during their first year or so of CERN. delivered at the beginning of 1965. It life, but, despite the problems, physi­ By mid-1963, the 709 was saturated was one of the first of the series and cists quickly became addicted to and was replaced in September of problems came with it. The effect on computing and could not get enough. that year by an IBM 7090 which gave the Laboratory's computing was eased The increase in workload from another increase of four in computing by using outside computing facilities bubble chamber and electronics ex­ capacity. This changeover was pain­ but the development plans for com­ periments at the proton synchrotron less for the users as the new machine puting services around the 6600 were early in 1960, meant that the Mercury was completely compatible with the delayed for at least two years. was saturated before its successor 709 and no programming changes This computer also brought the arrived. This was an IBM 709 which were required. It also involved no wonderland of multiprogramming as provided a four-fold increase in com­ change in the mode of operation, the next stage in the continuing battle puting capacity. Since it was a well one program completed its comput­ to bridge the gap between computing established machine, CERN was ing cycle before another started. and input/output speeds. In order to provide increased capacity and a more reliable service, the CDC 6400, a smaller compatible brother of the 6600 was installed in April 1967. About the same time manufacturers were asked to send information on their future products and technical discussions were started. CERN estimated its future needs as a sys­ tem which had a potential capacity by the years 1974-75 of ten times that then installed and the discussion revealed that several firms had plans to make machines larger than the CDC 6600. However the possibility of getting a new, large, well-proven ma­ chine by 1970 was very unlikely and CERN obviously wanted to avoid any repeat of its 6600 experience. A fur­ ther conclusion from the studies was that the new system should be based on two compatible machines with an interval of two to three years between their purchase. Thus it was decided to go for an interim solution, either by extending the CDC 6000 system or CERN 366.8.69 62 CERN's new central computer, a CDC 7600, was flown into Geneva airport mid-February and can be seen being unloaded from the plane and being wheeled across the tarmac.

by choosing a medium-sized machine from another manufacturer, the me­ dium-sized machine to be compatible with a larger .machine. The first solu­ tion was adopted and at the end of 1969 the CDC 6400 was upgraded by increasing the memory and adding a second processor, converting it into a 6500- Extra disk and drum capacity were also added. In the summer of 1969, there was a further investigation to see what large computers would be available for delivery by the end of 1971. The most economic solution proved to be a CDC 7600 which has a computing capacity about five times that of the 6600. There is still no computer on the European market providing com­ parable capacity. With the slowing down of the expansion of Laboratory I since the construction of the big new accelerator was approved, it is estimated that CERN will need a second 7600 in 1976 rather than 1974 or 1975. (Laboratory II, incidentally, is CERN 165.2.72 a customer of Laboratory I for its remote self-operated input/output sta­ nals are connected to the CDC 3100 large-scale computing). Overall, the tions (RIOS) as well as by card via a Hewlett Packard 2116B demand for computing is continuing readers and line printers located at computer. to grow but the rate of growth has at the central computers themselves. In An interactive display is attached least slackened off from a doubling addition, there is a car delivery to the 6600 via a Ferranti Argus 500 every year, which was the case service to a number of remote parts of computer and is used for interactive through to the mid-60s, to a doubling CERN. work which requires the computing about every two years. Semi-interactive facilities are avail­ power of the 6600. For graphics work able on the FOCUS system which is requiring less computing a CDC 3200 The Present implemented on a CDC 3100 with is available which has a large CDC channel-to-channel connections to the interactive display attached. It is used The present central computing ser­ CDC 6500 and 6600. FOCUS was devel­ particularly for the re-measurement of vice operates 24 hours per day, seven oped to provide economic facilities for bubble chamber events which have days per week including most holi­ quick sampling of experimental data been rejected by the off-line chain days. It is still predominately ' batch- (collected by small process com­ of programs. oriented ' with the main programming puters) at the central computers. Thus Over the last two years or so an languages being FORTRAN and the 3100 has direct data-links to the interesting change has taken place assembly language. A very large pro­ experimental halls. It has much wider in the manner in which many users gram and subroutine library is avail­ application, however, since it allows run their jobs. With the introduction able on disk in re-locatable form, and about twenty users simultaneously to of FOCUS on a wider scale, a number a tape library of some 35 000 labelled manipulate files at terminals and of users changed their working habits. tape reels is maintained close to the transfer job input files to either 6000 Keeping their programs in the perma­ two computers. machine for processing (with priority nent storage system of FOCUS, chan­ Batch input/output using card if appropriate) and receive job output ges were made directly to this readers and line printers is via three files back at their terminals. AM termi­ version rather than to the card deck.

63 The CDC 7600 'main frame' being installed at CERN. It will increase the computing capacity, in comparison with the 6600, by a factor of about five.

With direct access via the 3100, pro­ memories, one of 65 K words of 60 jobs is done exclusively using a grammers could obtain a larger num­ bits plus 5'parity bits, and the other device with a very high instantaneous ber of 'debug runs' than was possible of 512 K words of 60 bits plus 4 parity transfer rate (about fifty times that with the operator input/output system. bits. Programs are executed in the of the tape units). In fact a sequential Also a number of new applications smaller memory, but can transfer infor­ scan of a large file on the 7638 is very became practical — for example, mation between the two cores at a fast, a complete reel requiring about maintaining equipment inventories rate in excess of 36 million words ten seconds to read through. which required frequent changes and per second. Input and output oper­ The peripheral equipment attached periodic listings. ations are carried out via small peri­ to the 'front end' 6400 reflects the The addition of the remote input/ pheral computers with access to part way computing is expected to evolve output stations which took place in of the small core memory. at CERN (and elsewhere). The major 1971 brought about a further major The problem once again, however, items are : change. The RIOS eliminated the is to feed the information between High speed 9-track units for densities delay introduced by waiting for oper­ the processor and the outside world up to 1600 bpi illustrating the conti­ ator intervention and users can now at a sufficient rate since, in general, nuing trend towards higher densities obtain many more runs per day. the peripheral equipment is very simi­ (the present units have densities up Within a very short time a substantial lar in speed to that of the 6600 (in to 800 bpi) ; proportion of the total number of jobs some cases it is the same equipment). Multi-spindle disk drives to provide was being fed to the 6500 and 6600 The one exception is a very fast fixed permanent file storage ; via FOCUS and the RIOS and already disk (a 7638) with a large capacity Remote input/output stations based on fifty percent is input by the user him­ and fast transfer rate. Thus CERN has Computer Technology Model 10 Satel­ self. The number of short jobs has decided that the only peripheral lite One computers consisting of rapidly increased, at present nearly equipment attached directly to the processors with 8 K, 1.5 core 9000 jobs are processed each week, 7600 shall be two of these disks. All memory, card reader and line printer. needing 5000 tapes to be mounted. other peripheral equipment will be Five of these RIOS will be installed attached initially to a 'front-end' CDC initially and are scheduled to be oper­ The Future 6400 computer which is connected to ational around the middle of this year. the 7600 by a high-speed link. As a Another five will probably be installed On 15 February the new central com­ result all program input/output is be­ later in the year. It is interesting to puter, a CDC 7600, was shipped to the tween large core buffers and the note that, although once again CERN site. The future of large-scale com­ 7638s. Thus all files must be created was obliged to buy its large computer puting at CERN for the next few years on the disk whether they be ultimately from the United States, the best value centres around this machine which cards, printed output, magnetic tape for money for RIOS was provided by is expected to be operational by the input and output or permanent files a European manufacturer. An impor­ end of March. The arithmetic power residing on a multiple spindle disk tant criterion in making the selection of the computer is such that it can drive. was the expandability of the stations perform an addition in 110 ns, and a Copying between the 7638 disk and since there is a clear need to attach multiplication in 137.5 ns. As with the the external media will be taken care other peripherals, including tape units, 6600, many instructions can be in of by the system and is called stag­ to them in the near future. This process at any time. The effective ing'. This is not new for card input/ implies increasing the speed of the rate of executing instructions is output and printed output (it is used link between the station and 6400 ; around 20 million per second. It is on most large computers including Card Reader, Card Punch, Line Prin­ instructive, confronted by these figures the 6000 machines) but it is new (and ters to provide the usual peripheral which represent one of the most unique) for complete magnetic tape facilities. However, the volume of out­ advanced computer systems now reels or disk files. Only the very large put generated will be so high that available, to remember that just capacity of the 7638 disk makes it it is planned to attach a high speed thirteen years ago with the Mercury possible to include magnetic tapes in alpha-numeric microfilm printer at the 'add' time was 180 ^s and only the staging philosophy - one 7638 disk the beginning of 1973 to handle a one instruction could be in progress can hold about fifty complete tape substantial part of it. at any one time. reels. The advantage of staging every­ In comparing the advent of the The 7600 has two ferrite core thing is that input/output for active 7600 with that of the 6600 it is impor-

64 new files, and submit jobs to the batch processing system of the 7600. Output files may be retrieved at the terminal for examination. INTERCOM does allow jobs to be run on the 6000 machine with a measure of inter­ action and this will be exploited for CERN's interactive work, for example the GAMMA system (see the articles on interactive computing). Interactive work will be able to use a large core memory of 500 K words which will be CERN 200.2.72 shared between the 6400 and 6500. In tant to note that the serial number of can be carried over easily from one particular, it provides a very conven­ the 7600 is 19, whereas that of the CDC version to another. ient backing store to which programs 6600 was 3. Thus there are already An extensive hardware and soft­ can be swapped. many 7600s in operation (all in the ware acceptance test was carried out While the permanent file space will USA with the exception of one in the succesfully in CDC's factory in Min­ be adequate for storage of program UK which is at ICL's factory for inter­ neapolis running CERN programs. and small data sets, it falls far short facing to an ICL 1904A as a front-end; This is being repeated at CERN prior of the capacity needed to substantially others will be installed in Europe this to having a thirty-day live' acceptance reduce the use of magnetic tapes. It year). The first 7600 has been operat­ period during which users will run is probable that by 1974 the compu­ ing for a customer for two years. Also, their normal work on the machine. All ting centre will have around 70 000 CDC have obviously used their expe­ the evidence to date suggests that magnetic tapes. As a first step to rience with the 6000 series of machi­ CERN's 7600 is one of the best. increase the capacity of on-line stor­ nes in the design and manufacture of The disadvantage, from the relia­ age it is planned to replace the the 7600. Thus the memories now bility point of view, of channelling all multi-spindle disk drives by larger have parity bits and a lot of thought input/ouiput via a front-end machine capacity units which will become has gone into the diagnosis and is obvious. To overcome this the CDC available at the end of this year. repair of failures. The 7600 is fitted 6500 will be shutdown in the Autumn However, this will only 'buy time' and with a maintenance control unit in and modified in readiness for attach­ CERN is investigating the addition of the form of a peripheral computer ing to the 7600 as a second front-end. a 1012 bit data bank about the end which monitors the 7600's operation The modifications include adding ten of 1973 (see the article on data stor­ and diagnoses malfunctions. To repair more peripheral processors and age). This data bank will be acces­ a fault normally consists of replacing twelve more channels to bring it into sible by a number of computers and the defective subassembly and, up to line with the front-end 6400. Peri­ form part of a network of computers now, average repair time has been pheral equipment (including that at which is being studied. less than two hours. present on the 6500) will be allocated Quantitative performance analysis The software of the 6600 (or rather to one or other of the 6400 and 6500, of large systems such as the 7600 lack of) was a major headache for but, if either fails, its equipment, with the long term aim of perfor­ CERN and for the 7600, while the including RIOS and terminals, can be mance prediction is an area of increas­ position is very much healthier, it is switched to the other machine. ing importance since intuitive ideas more of an unknown than the hard­ Interactive facilities will be provided are often wrong. CDC has developed ware. CERN will be one of the first using the subsystem INTERCOM on the a special hardware monitor for CERN users of the 7600 software called front-end machines. CERN's require­ which will be used to study the flow SCOPE 2.0 (the software in the front- ments here are, primarily, on-line of work through the system. end is modified SCOPE 3.3 which is file editing and job submittal in a As mentioned earlier the 7600 is well established as the normal CDC batch processing environment. INTER­ installed in a new computer centre. 6000 operating system). It has been COM will allow users to have access The building consists of two parallel decided to use standard CDC soft­ to files stored in the SCOPE perma­ concrete wings, where there are offi­ ware making only such changes as nent file system, modify files, create ces and technical equipment, which

65 CERN 374.2.72 Installation of the 7600 well advanced in the This splendid new building is now everywhere, towards 'distributed com­ large hall of the new computer centre. Top left, the shape of the main-frame can be being progressively occupied by the puting ' — that is, having computers picked out. Then comes the 6400 computer, staff of the Data Handling Division. individually assigned to specific func­ the large core memory which will be shared tions rather than trying to use one by the 6400 and 6500 (Extended Core Store) and, nearest to the camera, a number of controllers large central installation to cope with which connect peripheral equipment to the 6400. The smaller computers everything. Examples of this are the On the right and in the foreground are the Omega and Split Field Magnet com­ operator displays. So far we have concentrated on the puter system (described later) centred glamorous super-computers which on a CM 10070 and the ERASME com­ cater for the large-scale computing puter system (also described later) enclose the computer hall proper and needs of the Laboratory. A recent centred on a PDP-10. In the electro­ support its roof. This hall has an area survey identified 80 other computers nics experiments field and the bubble of 1600 m2 and its height between false sprinkled around the CERN site. They chamber film measuring field, respec­ floor and false ceiling is six metres. are from fourteen manufacturers ; 59 tively, these are amoung the most There is also a basement which is of them belong to CERN and 21 have advanced of their kind in the world. occupied partly by air-conditioning been brought in by visiting groups Nevertheless, apart from the big equipment and electrical generators from research centres in the Member dedicated computers there is growing and partly by computer supplies. In States. Many of the machines are need of data links between compu­ particular, the bulk of the tapes will quite small, such as the PDP-8, ters. Higher data taking rates and be stored directly under the area PDP-9, PDP-11, PDP-15, DP516, larger total volumes of data to be where the tape units are located. A HP2116, IBM 1800 but they range as analysed are expected from the detec­ paternoster is installed for the physi­ large as the CDC 3100, IBM 360/44, tion systems coming into use. This cal transfer of the tapes. CI I 10070. can only lead to even more extensive The equipment in the hall is ar­ Almost all of them (with the use of computers in real-time, requi­ ranged by function, therefore there is exception of an IBM 360/30 used in ring either increased computing capa­ one area for input, one for consoles, administrative data processing and a city or more data links to a medium etc. Part of the hall is taken up by PDP-11 coming into use for library or large multi-programming computer. a 'Cceur Central' which houses all the needs) are involved in the physics There are already four data-links into user facilities. This area has two programme. Examples of their appli­ the central computers via the FOCUS floors, the ground one will house a cations are — on-line acquisition and system. A considerable number of remote input/output station and gener­ control in electronics experiments, on­ the 80 computers will require links in al input/output facilities. Cards, prin­ line data acquisition and control in the future to larger machines. ted output, etc. will be physically scanning and measuring system for The following articles go into the moved between the Cceur and the bubble chamber film, interface and various computer applications in more hall via conveyor belts. The first floor file-handling between a number of detail. houses the Advisory Services, a pro­ computers, control of accelerators grammer work area, a conference and beam transport systems, etc. room and a visitors' gallery. There is a strong trend, noticeable

66