Jump Over Left Menu
Central Computing Annual Report Excerpt 1986
Advanced Research Computing: The Cray X-MP/48
This year's most significant event was the arrival of the Cray X-MP/48 supercomputer which SERC will operate on behalf of all the research councils and the British academic community.
In August 1985 the Advisory Board for Research Councils, the Computer Board and the University Grants Committee Joint Working Party on Advanced Research Computing, chaired by Professor A.J. Forty produced a report which found that there was a strong case for new advanced computing facilities. It received wide coverage in the press. The report advanced convincing reasons based on needs in all fields of research and development.
It recommended that ABRC, the Computer Board and UGC should secure and allocate funds for a national facility for advanced research computing. This should comprise:
- a central installation, at the Atlas Centre, RAL, consisting of the most powerful supercomputer available.
- an enhancement of JANET - the Joint Academic Network to ensure good communications for remote users.
- a distributed system of other forms of advanced research computing.
- a national organisation of advanced research computing to ensure the effective use of these facilities, to encourage collaboration of all kinds, including industrial and overseas collaboration, and to stimulate new developments in this important area.
Given the scale of the recommendations, there has been remarkable progress toward their implementation during the past year. In February the ABRC, Computer Board and UGC agreed on a funding package which would provide a Cray X-MP/48 supercomputer to be installed at RAL and a less powerful Cray IS supercomputer to be installed at the University of London Computer Centre. ULCC's Cray 1S came into service in Autumn 1986.
Cray's X-MP/48 is one of the most powerful computers obtainable. It could perform about a thousand million arithmetic operations per second but this theoretical performance is not realistically attainable. However, careful programming and optimisation of existing software allows a useful fraction of the machine's power to be applied. This is considerably more than can be obtained from conventional mainframe computers.
The computer achieves its high performance in three main ways. First, it is constructed from very fast components assembled compactly. It also contains special hardware that handles vectors in the same way a conventional computer handles numbers. This can provide large gains in performance. Finally, the Cray contains many functional units which can operate simultaneously, and clearly the more this parallelism is exploited, the greater the overall performance.
Preparation and installation was a major undertaking. Initial planning started in January, with a series of preliminary meetings with Cray personnel to establish the needs of the machine in terms of power, cooling and floor space, and the needs of the considerable supporting plant.
In April, the Cray Installation Working Party held its first meeting. Its brief was to direct and oversee all aspects of the installation, to ensure that all proposed uses and applications of the X-MP/48 were catered for and that installation milestones and targets were achieved. The IWP met on a further seven occasions and was disbanded on December 12 following the completion of a successful programme.
Wednesday December 3 saw the arrival of the machine at the Atlas Centre. It was first powered up on December 5 to start the engineering QA testing and commissioning period. RAL acceptance testing will follow immediately in early January 1987, and the full service to users starts on February 6 1987.
This Advanced Research Computing facility has its own Management Committee. Applications for use of the facility are made through the normal peer review procedures, although Director, Computing has authority to approve small amounts of resource for specific projects to enable feasibility studies.
The processor complex of 4 CPUs and 64 Megabytes of memory is the most powerful in the UK, and attaches to 14 Gigabytes of high speed disk storage and 256 Megabytes of solid state fast secondary storage. It runs the Cray Operating System (COS).
The Cray is run as a batch job system. Users can submit work directly from the RAL interactive VM/CMS system, and from JANET nodes. Output can be routed anywhere on JANET. Both FTP - File Transfer and JTMP - Job Transfer and Manipulation protocols are supported.
Several applications packages and libraries are installed on the Cray, including the GKS and GINO graphics systems, and many programs from the user community in areas such as quantum chemistry.
Earlier this year, the various accounting, rationing and tape management systems on the RAL mainframes were redesigned, simplified and rationalised. This effort has paid off, allowing the Cray operating system to be connected to the new unified Atlas Centre mainframe accounting and rationing database with little effort. The Cray system is also ready to be connected to the Atlas Centre tape management database, allowing allowing direct use of users magnetic tapes from Cray jobs.
VM/CMS: The Interactive Service
VM/CMS now has over 2,000 registered users, and a peak of around 380 active users on most days, this demand is catered for by a hardware upgrade from the IBM 3081 D to the 3081K with an increase in CPU power of 50% to around 15 Mips.
IBM software for these systems has remained stable for most of this year, with VM at Release 3.2 and CMS at Release 3.14. Preparatory work has been undertaken for moving to higher releases in the new year (VM 4.2, CMS 4.13), and development of local software has provided greater flexibility in the system available to the general user.
Management of CMS minidisk space has been automated using a locally written service machine, Stager. This regulates space allocation, backup and restore of minidisks and provides a powerful access control facility for generalised resources (including minidsks, spool files, and privileged commands). Much work has been undertaken to drive the backup/ restore function into the M860, a tape cartridge mass storage device.
A major initiative to unify system-wide database functions as applications of SQL/DS (the IBM database product installed last year) has been undertaken. As a result the restructured VM accounting system is already in production, designed to allow future incorporation of MVS and COS accounting, and to reduce the effort expended on accounting by both users and CCD staff. Work is also underway to unify the structure of the tape management system.
MVS/JES3 : The Batch Service
This year saw the introduction of additional batch processing capacity with the installation of the local MVS system in the front-end IBM 3081 machine supplementing the MVS capacity in the Atlas 10 computer. This allows the Central Computing facility to deliver additional batch computing resource overnight.
MVS has also been affected by the changes to using a unified front-end database system for accounting, job status information and tape handling as explained above.
Upgraded software for access to the Masstor M860 mass storage system device and for access control and security has been installed.
In mid March a 42in Electrostatic Colour plotter was directly connected to a channel on the mainframe via a model 830 RPM rasterising controller and is accessible from JANET. This is a high definition device capable of 40,000 dots per square inch and supports 256 colours plus greyscales. It is used in much of the chip design work. Full colour output is produced at approximately five minutes per metre.
Paradyne Pixnet Channel Extender
Pixnet provides the ability to extend mainframe Channels supporting certain peripherals over great distances while maintaining machine room response times.
In August a Pixnet system was installed between RAL and Central Office Swindon. Attached to the JACS IBM 4381 and RAL's Central mainframes, it allows users of IBM 3270 style terminals at either site to access either mainframe. A cabling exercise is continuing within CO to make this facility more widely available there. This facility will be used particularly by the Central Office users of PROFS.
Amdahl 4705 Communications Controller
The Amdahl 4705 communications front-end processor came into service in June after serious software problems were solved, and the three Memorex 1270 communications controllers which it replaces were finally removed. Two additional 32-port IBM 3274 screen controllers were added during the year to cope with increasing demand for the PROFS service and to provide extra resilience.
Account Units Used -1986
Use of RAL Mainframe Computers by Board
Computing Infrastructure VAX Systems
Two extra megabytes of main storage were added to the VAX 11/750, doubling its capacity. RAL computing staff now provide support for over 50 machines. This covers Digital's networking software and the coloured books software produced by UWIST and St Andrews as well as JTMP software produced by UWIST. A central network support contract for these software packages has been organised for SERC-funded VAXes. So far 47 SERC-funded VAXes have joined the contract.
A MicroVAX II has been purchased. This will offer a VAX Front End service to the Cray X-MP/48.
Progress on GIFT - the General Internetworking File Transfer project continues. The first stage, which allows JANET users to initiate file transfers from their own computers to and from computers on CERNET and DECNET at CERN, has been upgraded to run under the latest release of the VMS operating system and Digital's networking software. Further development and testing has continued on the second stage which allows similar file transfers to be initiated by CERNET and CERN DECNET users.
Wide Area Networking - Joint Network Team (JNT)
JNT - the Joint Network Team, is responsible for creating a unified networking infrastructure for the universities and research councils.
This year saw the continued growth in the use of services based on the interim Coloured Book standards. The name registration scheme now lists over 700 systems with networked terminal access, 583 systems supporting networked file transfers, 633 mail systems and 152 systems with job transfer. Two other initiatives, the support for networked screen-mode terminals and the networking of personal computers, have matured considerably. There has also been progress in the provision of products conforming to the Pink Book Standard for Ethernet.
The network installation programme was expanded as a result of new funding awarded by the Department of Education and Science in connection with the Forty supercomputing initiative. The new funding will be used to upgrade the performance of campus networks to complement the performance upgrade planned for JANET. In addition, a major project to upgrade the performance and functionality of the Community's existing PAD population is in progress.
The task of defining a strategy for the transition from Coloured Book standards to the new ISO - Open Systems Interconnection standards is nearly complete. The group responsible for this work published a draft strategy that has been widely circulated both within the community and external organisations. The group is now finalising the document, taking into account the comments that were received. The first step in the transition from Grey Book mail to X.400 has been taken with the introduction of a gateway to provide access to the growing number of X.400 mail systems outside the Community.
Substantial support was provided for the European RARE initiative. The team increased its participation in national and international groups defining ISO standards.
The network executive is responsible for the operation of the JANET network. It does this for the Secretary of State for Education and Science, who is advised on such matters by the Computer Board for Universities and Research Councils.
This year saw the consolidation of JANET, with emphasis placed on improved reliability. Significant progress has been made, although the MTBF of a switch is still much lower than the three months planned for. During the year, the processors at ERCC and Daresbury were upgraded from GEC4160s to GEC4190s, giving a potential threefold increase in capacity at these sites. Several more trunk connections were enhanced to 48kbps capacity. Finally, the Bidston and Swindon switches will be phased out at the end of March 1987, and a programme of re-assigning the sites supported by these switches to other switches in the network is well advanced.
Detailed analyses of traffic and performance are being done on a regular basis. Traffic levels have grown around 50 percent in the last year, and by early December the network was handling some 650 megabytes of user traffic every working day.
The Executive plans to upgrade the central core of links on the network - those between RAL, Daresbury, University of Manchester Regional Computer Centre and University of London Computer Centre, to 256kbps during 1987. It also expects to improve site access for the present norm of 9.6kbps to a norm of 48kbps or 64kbps over the next three years. This has been made possible by a substantial injection of additional funding from Department of Education and Science.
Information Management Services
IBM's Professional Office System - PROFS, now has over 500 users. Plans to extend the service to Central Office are at an advanced stage. During the year the new release, PROFS Version 2 Release 2, was installed, including all the modifications and additions required for the RAL environment - especially in the area of networking. RAL is now a reference site for PROFS and we receive many visitors wishing to see our implementation. Also, computing staff from RAL have been very active in the EPUG, the European Profs Users Group, which resulted in many requests for advice and assistance.
Status - the AERE Harwell-developed package for free-text handling was released to customers in its VM/CMS version.
The RAL-developed R-EXEC system is now heavily used on APS board projects, notably AMPTE and projects within the remit of the Geophysical Data Facility and the World Data Centre.
SQL/DS, the IBM database product, is heavily used for administrative and scientific support systems, within user-defined applications developed by Information Management Group of Central Computing Division. In addition to the systems developed for control of and information on the central computing operating systems and their resources, particular systems developed during the year handle stores and control room information for ISIS, dosimetry information for Health and Safety, accommodation booking for visitors to RAL while developments have continued on FDS (Financial Data System) and the database of users of RAL facilities.
Computing staff continue to work on the appropriate standards bodies including acting as BSI representatives to certain standardisation meetings. This work results in better standards, and more relevant standards, for these areas of activity.