As expected, fewer Cray users attended the second meeting than attended the first. However, with about 30 users present, some having travelled from as far afield as Belfast and Strathclyde, most of the scientific disciplines were represented.
The meeting was opened with a brief welcome by the Chairman, Professor Geoff Pert.
One of the changes since the last meeting, and one which affects all users, is the change which has been made to the charging algorithm. We need to make efficient use of the Cray resources and the new algorithm encourages this.
It has been suggested that a central user subroutine/program library is created. If you think you have a routine or program which other Cray users might find useful, please contact Alan Mayhook, who will arrange for it to be installed and indexed. Sorry, but no maintenance on these routines can be offered!
The FORGE system - a set of optimising tools which runs on a SUN workstation has been installed. Unfortunately, the cost of a multiple machine license is prohibitive, and a single machine license has been bought. This machine resides at the Atlas Centre: users are welcome to visit Atlas and try it out on their own code - just contact Roger Evans or Alan Mayhook.
Talking of SUNs. a UNIX Front-End has been mounted on a SUN workstation. Roger would like to hear from any interested user wanting more information.
GKS version 7.4 will be mounted on the Cray at the same time as it is mounted on CMS (no date given); eventually, 7.2 will be withdrawn. However, version 7.4 is already available for testing.
The limitation on the size of plots for the Versatec has been lifted. Now, up to at least 40 Mbyte files can be plotted.
The DIMFILM package has now been installed, to enable users to generate output to be processed into film on the DICOMED film recorder at ULCC. It currently writes to a single magnetic tape only, a problem which prevents multiple tapes being used is being investigated. [This problem has now been fixed - CDO.] The tapes are sent to ULCC for processing, the turnround can be quite long (about a week). It may be possible to send small files over the network and so speed up the turnround - anyone interested should contact Roy Platon
The IRIS 3130 Graphics Workstation has now been installed at the Atlas Centre. There is a wide selection of software packages available for this workstation - and again, users are welcome to visit Atlas and use it - just contact Roy Platon in Graphics Section to arrange everything.
The UNIRAS set of packages is scheduled to be mounted this year. Looking at 3D graphics packages, there are several possibilities including OASIS, DORE and MOVIE-BYU - all with strengths and weaknesses. Chris Osland would like to hear from users about the type of graphics they anticipate they will be producing, so that a sensible choice of graphics package can be made.
COS: Our current Cray Operating System, is now running at level 1.16.1. with FSS pre-emption active. The next bug fix level (1.16.2), currently being installed, offers improvements such as better tape support, a range of symbolic debuggers, and an improved GOS feature. GOS is the Guest Operating System - which allows a single UNICOS guest system to run under COS. COS 1.17, scheduled for October 88, offers even better GOS support, but will only support UNICOS release 4.0 and higher.
Cray will make no further development of COS after level 1.17, and there is already Cray hardware for which only UNICOS is available.
UNICOS: has become Cray's strategic operating system, for the future. Release 3.0.8 is now being actively installed as an initial version for development, and this will be mounted under GOS as soon as work on it and COS 1.16.2 is completed. Release 4.0 is scheduled to become available in July 88, but is dependent on COS 1.17 (available October 88). Release 5.0 is currently scheduled for availability in first quarter 89, and is the first to provide COS functional equivalency. It will be the earliest release on which a full user service could possibly be offered.
UNICOS currently seems to be the best future operating system option for Cray hardware at JRCSU, and on this basis users are encouraged to become involved in the preparations for migration as early as possible. A series of user notes, courses, conversion documents and tools is being planned, and any queries or input which users wish to make at this stage should be directed to the Systems Group via User Support.
At very short notice. Professor Bilby and Dr Cardew gave an interesting talk, which many users obviously appreciated. Their work involves the study of crack propagation using Finite Element methods, and having discounted the use of ABAQUS. they wrote their own program, with additional features and the ability to upgrade the program easily. Their main problem is simply one of huge data files (I noted several heads nodding in agreement on this point). They actually use an "ancient PERQ" to display graphics. This can take up to an hour to load a typical 3 Mbyte file, but the resulting display is "quite impressive".
They once tried to run a large job rather than one with lots of restarts. This job was put in requesting 80,000 seconds (on one processor). However, 2 days later, the Cray crashed, and the job lost - this is not an experience they have repeated! Yes - they agree that the code of their program could probably be "improved", but they don't have the resources to do this. After all, they are in the business of scientific research, not computing, and they need to produce results (again, the nodding heads).
CFT 1.15 BF2 (Bug Fix 2) is now the production Fortran compiler. The alternative compiler. CFT77 2.0 is on test - and seems to be an improvement, producing better, faster code in most circumstances. C 3.0 has been installed LISP and PROLOG are available and Pascal 3.3 is on test. Moving on to libraries, NAG MK 12 is now in production, with new routines and support for Fortran 77. CERNLIB, with its general data manipulation routines is also available. These routines have not been "vectorised" much yet. but now that CERN also have a Cray, this is likely to happen.
Other packages and libraries which are either on order or have been recently installed include GAUSSIAN 86, ABAQUS, PAFEC, TOSCA FELIB and LUSAS.
A Central Tape Database is being developed which will be accessible from Cray as well as other systems at Atlas, in both batch and interactive mode, and is now in its "user consultation" phase. A paper describing the user commands to be available was distributed. Users were asked to digest this and pass comments to John Gordon. The paper is available in CMS (type FIND TMS WRITEUP). The new 3480 tapes - small cassette-like tapes - hold about the same data as the magnetic tapes of old. With improved technology these tapes are likely to hold up to 1 Gbyte in the future. Anyone likely to use magnetic tapes should consider 3480 tapes. Please let User Support know if you would like to try out these tapes as a Cray service is being planned.
Finally, a plea for feedback from Cray users: let User Support know if you want:
This is an extra item, not on the original agenda. Dr Brian Davies took the opportunity to impart some new information. As most users will already know, the IBM 3081 and the Atlas 10 were replaced by an IBM 3090 200E. This machine has two processors, runs at about 30 mips, and has a Vector Facility capable of a peak performance of 116 Mflops ("peak performance" - the performance guaranteed that no user can exceed!).
At the beginning of May, IBM announced that the Atlas Centre had been chosen as a third member, to participate in their European Supercomputing Initiative (following CNUSC at Montpelier and CERN in Geneva). The program aims to develop partnerships between IBM and academic institutions specialising in numerically intensive computing techniques. It is hoped to be able to upgrade from the current 200E to' a 600E - with 6 processors and 6 vector facilities - but when and how is not yet clear.
One possible scenario might result in the additional capacity being shared between existing SERC-funded work, industrial partners of RAL and ABRC-funded work. This could have a benefit for Cray users. The IBM 3090 has faster scalar processors than the Cray so the significant amount of work now running on the Cray that does not fully exploit the vector hardware could run faster on the IBM, thus freeing capacity on the Cray for work that does exploit vectorisation.
The IBM service is seen to complement the facility offered by the Cray - users interested in finding out more about the facilities offered on the IBM should contact User Support.
Several points were raised in this session, I have listed them below.
Examples of jobs - users would like to see "recipes" for different types of job. The set of simple examples handed out on the introductory course needs to be extended. Users are urged to suggest topics to User Support so that more examples can be included in this set.
Machine limitations - there is a maximum of 300,000 Cray blocks for a dataset (this is 1.2 Gbytes - more than a tenth of the total disk space available). Jobs fail if an attempt is made to create a dataset larger than this.
Migration of permanent datasets - this should only happen if a dataset has not been used for 2 days. However, on Monday morning (following the weekend), users find it takes ages to recover their datasets, because they have been "migrated" to the Masstor, and everyone is trying to retrieve them. This should be fixed soon.
It was emphasised that users need to make it known what needs they anticipate for 3D graphics. The problem is that users don't use a 3D graphics package, because one doesn't yet exist on the Cray - but if one did - what would users expect it to provide?
If finance were available for an upgrade, users were ask to discuss the possibilities. Extra memory would be very useful, the Cray seems to be under-resourced in this area when compared to the power of the machine. However, a balance needs to be be kept between the size of memory and the size of the SSD. Cray disk space is another limiting factor which users would like to see resolved. However, the capacity of the permanent disk filestore is governed by the size of the Masstor, where most datasets reside for most of the time; the size of the disk space really just governs the amount of "temporary" disk space. Although use of the Masstor by MVS is decreasing, there is an equivalent increase in the use made by CMS as users move their jobs to SLAC Batch. The new 3090 will also make use of the Masstor. By the end of the year, if the Masstor is not increased, and if users of the Cray continue to increase, the disk allocation may need to be cut.
In summary, an upgrade to the VAX front-end was seen to be the least important of the possible areas for enhancement; with more memory and SSD more science could be achieved; with extra disk and Masstor a great inconvenience would be removed - and most users would benefit. However, before any enhancements are made, they have to be financed!
Finally, a plea from us. Before upgrades to the Supercomputing facilities are agreed, the first question that will be asked is "what use has been made of the current facilities?" A letter is being sent to all grant holders, requesting users to supply information on the scientific achievements that have been made so far. A page or two from each user will suffice - along with copies of any reports which have been written. These will then be edited into a "book of achievements". In order for us to do this editing in time for the first of the relevant Committee Meetings, we need this information by MONDAY 1 AUGUST.
