In response to requests from users at previous meetings this meeting was held in London, at the University of London Computer Centre. About forty people attended and we were delighted to have Roy Baker from ULCC and John Clegg from UMRCC present.
Following a brief welcome from the Chairman, Professor Pert, the morning session consisted of presentations from staff at the Atlas Centre and as a new feature, from the other two National Centres.
Roger Evans gave an introductory talk covering subjects that would not fall into the more detailed talks later in the morning. It seems to be possible to provide an upgrade on the current Micro VAX front end service which has proved to be very popular with users. The current Micro VAX 2 has limited CPU capacity and its performance will reduce if the workload increases much beyond the current level. It is intended to purchase a Micro VAX 3500 for the Cray front end service and sell on the Micro VAX 2 internally to the VAX development group. The 3500 has about three times the CPU power and should be satisfactory for the life of the Cray X-MP/48 service.
The IBM 3090 front end now has an IBM vector processor attached, as had been announced by Dr Davies at a previous meeting. The maximum vector speed is only about half that of the Cray X-MP but scalar performance can be comparable. A few benchmarks have now been run, confirming the vector capability but showing that, particularly for handling large matrices the scalar performance can drop to about two thirds of the X-MP due to cache "misses". On the other hand the IBM VS Fortran vectorising compiler certainly "tries harder" than CFT or CFT77. It is particularly good at vectorising outer loops or splitting long loops into vectorisable and non-vectorisable parts.
Users were reminded that some commercial software packages are very expensive and where possible their cost should be included on grant request forms. It is sometimes possible to arrange a trial period with the supplier on the understanding that if users wish to proceed to production work then a grant request will be made later.
Finally a very sincere "thank you" was made to all who had contributed to the "Cray Science Report" which was about to be passed up through the relevant committees. The document contained seventy pages of high quality scientific results, 151 publications have been produced from data obtained from the Cray X-MP/48 and the report was only an interim report from the first year!
David Rigby gave a report on items related to operating system software - see other articles in this issue.
John Gordon gave a brief presentation on new software packages recently mounted on the Cray. The chemistry packages GAUSSIAN86 and MOP AC are now available as is ABAQUS v4.7 (following licensing difficulties) and LUSAS, both finite element structural codes. Another chemistry package, AMBER is currently being installed. Negotiation with Harwell are taking place about the fluid dynamics code FLO3D and its associated package ENTWIFE. Another fluid dynamics code PHOENIX is in the process of being implemented.
John reminded users of the availability of the FORGE optimising tools on a SUN 3/150 at RAL and pointed out that this is currently being ported to VM/CMS so that it will be available to all users without having to travel to RAL. The means of full screen access to the IBM CMS service were also described since several users were having difficulty with this (more details from John Gordon if needed).
Roy Platon summarised the graphics news: RAL GKS v1.11 is now available on the Cray. This is the full ISO 7942 standard of GKS 7.4 but users wishing to take advantage of its new features will have to make small changes to their calls - for instance the "open GKS" call now has an extra parameter. This version supports a greater variety of workstations and supports the concept of targeted metafiles where the actual pixel resolution of the output device can be returned to the user code. It supports CGM metafile output which is appreciably more compact than GKS metafiles (typically only 20% of the size), three levels are available: character is the most compact and is best for network transfer, binary is slightly longer but optimised for speed of interpretation and clear text is longer still but may be edited with any text editor.
The Silicon Graphics Ins 3130 at RAL now has a CGM metafile interpreter and metafiles can be disposed to the Iris directly from the Cray. The Iris is a high resolution colour device and users are welcome to make use of it. A high resolution (300dpi) colour hardcopy device will be attached to the Iris as soon as funds permit.
The RAL video facility based on a Topaz UNIX machine and proprietary frame buffers and video controllers is now available for a trial service. Like the Iris it has a CGM interpreter and metafiles can be disposed directly from the Cray. The delivery of UNIRAS is imminent and it will be mounted on the Cray and IBM 3090, it will produce CGM metafiles in the next release (version 6).
Roy Baker from ULCC described the changes which were about to take place at ULCC. This was an "interim replacement" financed by the Computer Board but also aimed to reduce running costs. The two Cray 1-S machines are to be replaced by an X-MP/28 (ie. two processors and 8 Million words of memory). The new machine will have three IO processors, 4 Million words of buffer memory but no SSD. The DD29 disk drives currently in use will be replaced by faster DD39 drives fairly soon after the main installation is complete. The first Cray 1-S is due to move out on 19th December followed by the installation and hardware hand over of the X-MP/28 by 9th January 1989. Installation of the ULCC software and user data should be complete by 16th January.
ULCC also have a VAX cluster to provide a Cray front end service. A VAX 8250 supercomputer gateway will be linked to the X-MP and a Micro VAX 3600 will provide a powerful user service. On the Amdahl 5890 front end the intention is to move to VM/XA/SP2 as part of the common user interface policy.
The Masstor M860 at ULCC is to be replaced by a Storage Tek STK4400 automatic tape handler to provide an order of magnitude increase in filestore capacity.
John Clegg from UMRCC described the recent changes there, an Amdahl VP1200 had arrived on 5th September as a replacement for the Cyber 205. The VP1200 at the national centre complements a VP1100 at the Manchester local service, the VP1200 is about three times more powerful than the 205 and is more powerful than the ETA 10 model that was the main competition.
The VP1200 has 64 M Bytes of storage and a 14 nsec scalar cycle and 7 nsec vector cycle. At 571 Mflops it is comfortably the fastest single processor at any of the national centres. The machine runs VM/XA plus the Amdahl VP/XA extensions and supports some users directly on a TSO service. The route in for most users is via an Amdahl 5890, which front ends both the VP1100 and VP1200.
The machine has been benchmarked at up to five times faster than the Cyber 205, although some code that was very heavily tuned for the 205 and used 32 bit arithmetic only runs at the same speed. There is a Scientific Subroutine Library available and also an interactive debugger and interactive vectoriser. The VP toolkit includes an execution profiler similar to SPY on the Cray and the HSIO high speed IO library. Currently there is no implementation of UNIRAS on the VP1200 but this is being pursued.
With the coming of UNICOS it will be possible to have a much closer integration of graphics workstations and supercomputer. One user from SERC Engineering Board is having a SUN workstation linked in on a dedicated 64 Kbit/s line. If anyone would like to make a case for similar facilities based on improved science or productivity then it may be possible to "pump-prime" for a period of say, six months with loan equipment from RAL. Please contact Roger Evans if you are interested.
The FRAM collaboration from NERC have an enormous data problem with 30 Million words of data being produced per run. Optical disks could prove a more compact and convenient storage medium.
Following the previous Cray Users Meeting the priority order for possible upgrades was set as: top priority - upgrade to 16 Million words of main memory, second priority - upgrade to 256 Million words of SSD, thirdly - install VAX 8250 front end. Dr Farmer from NERC pointed out that one of their projects was currently unable to make use of a state of the art code since it required 256 Million words of SSD.
A plea was made for all the National Centres to agree a standard for graphics metafiles now that CGM was available.
A small working party from Reading, NERC and RAL will look at the need by the NERC users for large volume graphical output on microfiche or microfilm. A suitable device is expected to be available in mid 1989 but no funds are currently earmarked for it. Some users expressed concern over jobs submitted by FTP whose output was then lost. Advice was given to use JTMP in preference and for long jobs to dispose the output to a Cray dataset as well as to the remote site for additional security.
