Jump Over Left Menu
Issue 24: February 1993
Flagship Issue 24
- The MasPar Challenge
- New Year's Honour for Bill Trowbridge
- Free Video Time!
- Trainee Scientists
The MasPar Challenge
There is a lot of debate going on about the future of High Performance Computing being based on Massively Parallel computer architectures. To demonstrate the ease with which some massively parallel computer systems can be programmed, MasPar decided to throw down a Challenge and invite computer users in Universities and Research Councils to try the MasPar massively parallel system. The Challenge was to port an existing program to the MasPar. The program needed to be related to some research activity or application development but not necessarily a complete free standing application code.
A timescale of three months was set and in June 1992, a MasPar MP-1 with 4096 processors was installed at the Atlas Centre and connected to the JANET network. This system represents one of the smaller MasPar systems, and yet is capable of delivering 300 MFLOPS and over 6000 MIPS. The top of the range MP-2 with 16384 processors is capable of delivering 6.3 GFLOPS and over 68,000 MIPS.
The MP-1 configuration included the MasPar Programming Environment (MPPE), MasPar FORTRAN (a version of Fortran 90), MPC (a parallel version of ANSI C) and VAST -2 (which assists the translation of F77 to F90).
The Challenge was issued bye-mail and a direct mailing to university departments and people were asked to provide a two page project description.
Over ninety people made initial enquiries and in excess of thirty projects were submitted. MasPar machines have a multi-user capability, but to ensure that each person was able to get a fair share of the resources, only eighteen projects were selected for the Challenge.
The successful applicants were asked to attend a free one day hands-on training session, where they were provided with details on how to remotely access the MP -1 and given the opportunity to ask questions about the system and to take away the relevant documentation.
During the Challenge period, participants were able to call a help line for support and technical assistance. A closed user group was set up for the participants so that any requests for long runs on a dedicated machine could be mutually agreed. MasPar only had to act as a referee on one occasion.
Participants came from a variety of backgrounds and departments including NERC, AEA Technology and various university departments. Contents of projects included chemical substructure searching, Monte Carlo code for Nuclear Physics, Fluid Dynamics, Cellular Automata simulation and C++ language development.
After three months of hard work on the machine from the participants, the real hard work started when the papers containing the results had to be produced! Thirteen completed project papers were submitted for adjudication by a panel of experts from MasPar and RAL. The selection of the winning paper was based on a number of criteria including:
- performance attained
- relative speed-up
- novel implementation
- innovative algorithms
All participants were invited to a Symposium to present their papers and hear the results of the Challenge. By kind permission of RAL, the Symposium was held at the Atlas Centre in October. Twelve papers were presented and everyone had an enjoyable day. The judges had a very difficult task, but eventually a winner was chosen: the winner was a combined paper from Dr John Collins and Shane Sturrock from the Biocomputing Research Unit at Edinburgh University, on Fast Protein Database Searching. The abstract is as follows:
We have implemented the Best Local Similarity algorithm on the MP-ll04 in a prototype program, MPsrch. This has been coded very effectively in mpl (the ANSI C language with parallel extensions available on this SIMD multi-processor machine). The program runs with less than 4 Kbytes of memory per processor regardless of database size and achieves 95% processor utilisation.
MPsrch allows the user to choose any Dayhoff PAM scoring table and an appropriate gap penalty. After the search the user can also specify the number of alignments to display, or request display of those alignments above some specified score threshold. Alignments are always presented in decreasing order of score.
The longest protein in Swissprot (V 21), 5037 amino-acids long, can be searched against the full Swissprot database in 314 seconds, equivalent to 126 million matrix cell updates/sec. Program overheads reduce the performance on average length queries, eg a search with a 250 residue protein query takes 30 seconds. MPsrch is substantially faster than any commercially available implementation of this algorithm.
MPsrch was successfully demonstrated at the Computing in Molecular Biology conference in Chester, 2-4 September 1992 and a poster display was accepted along with an abstract which was published in the proceedings.
MPsrch will be installed on the MasPar at the European Molecular Biology Laboratory in Heidelberg, Germany. Also it will undergo a trial period on a demonstration MasPar at the Human Genome Mapping Project Resource Centre, Harrow.
Further improvements such as asynchronous loading of databases will be tried and extension to nucleic acid searches will be made.
The runners-up were Robert Brown and David Wild from Sheffield University who investigated various aspects of computer based chemical information systems. Ian Stephenson from the University of York, who implemented his Creatures simulator, together with Mike Ashworth from NERC, who implemented his 3D Hydrodynamic Shallow Sea Model, were joint third.
The Challenge could not have been completed without the participants and the Atlas Centre who enabled the machine to reside on their site for the duration of the Challenge. MasPar would like to express their gratitude to everyone involved.
Victor Knight, MasPar Computer Ltd
New Year's Honour for Bill Trowbridge
Professor C W Trowbridge, one of the world's leading experts on the computation of electromagnetic fields, was awarded an OBE in the New Year's Honours List for service to science.
Bill Trowbridge is a graduate of London University, a Fellow of the Institution of Electrical Engineers (UK), IEEE (USA) and a member of the Institute of Physics. During an outstanding career at the Rutherford Appleton Laboratory, he formed a group in 1971 which led the world in the use of computers for the simulation of electromagnetic fields. Bill Trowbridge's international reputation was further enhanced in 1976 by his initiative in the organisation of the first COMPUMAG Conference held in Oxford. This series of conferences, now held bi-annually at venues around the world, has remained the foremost conference on the techniques of electromagnetic computation.
In 1984 he became a co-founder and Chairman of Vector Fields Ltd, a company dedicated to the developments started at the Rutherford Appleton Laboratory, in the private sector. The company has since become a world leader in the technology, supplying software to customers in 22 countries and receiving the Queen's Award for Technology in 1992.
In 1992 Bill was appointed Visiting Professor at Imperial College of Science and Technology for two years and is currently Visiting Professor at King's College, London. In 1992 he was elected to the Royal Netherlands Academy of Arts and Sciences as a foreign member and also received the IEE Science Education and Technology Division Achievement Award for 1992.
Free Video Time!
In response to requests for access to the Atlas Video Facility from all sections of the UK academic community, the University Funding Council (UFC) has provided funds for a pilot scheme to provide this. Until now, access to the facility has been free only for users of UK supercomputing facilities and SERC users, other users paying a commercial rate. FOR THE PERIOD OF THE PILOT SCHEME ONLY, all UK academic users may use the facility without payment; after that the service will continue to be available to these users at a special academic rate. Supercomputing and SERC users will continue to have free access.
The Atlas Video Facility, located in the Atlas Centre at RAL, provides a powerful system for the production of animation sequences on videotape. Originally designed in 1987 and commissioned in 1988, it was considerably enhanced by the addition of an Abekas A60 video disk (holding 750 video frames) and a more powerful processor in 1989. It has been used to produce video sequences for oceanographers, atmospheric physicists, engineers, ecological modellers and laser scientists, amongst many others. The sequences have been shown at many conferences and have been seen on all UK broadcast channels and abroad.
Previous articles in FLAGSHIP (issues 3, 6, 7, 9 and 15) have described the facility and some of the videos made on it. These have all been made for supercomputing or SERC users.
The UFC, through the Advisory Group On Computer Graphics (AGOCG), have provided funds for a limited period (around three months) and a limited amount of facility time (around 80 hours). These resources are available to non-SERC, non-supercomputing, UK academic users for the making of videos. When the pilot scheme ends, such users will continue to have access to the facility at special rates. Use of the facility is accounted in terms of three factors:
- wall-clock time for which the user's job has exclusive use of the video facility hardware;
- programmer and consultancy time required to assist the user in producing the video;
- extra consumables beyond the default output produced by the facility (usually one Hi-Band U-Matic SP tape and two VHS copies).
The UFC wishes to encourage UK academic users who want to make videos to try their applications on the Atlas Video Facility. For some users this pilot scheme will allow them to see whether video is a suitable medium for their requirements; others should be able to make a complete video within the timescale of the pilot scheme.
Several different input formats are supported and these are documented in a User Guide that is being made available. In addition the staff at the Atlas Centre will provide advice on style and content, which can be crucial in the production of a watchable video. A Style Guide is being prepared, based on the experience of previous users. The completion of the documentation is another aspect of the pilot scheme funded by the UFC.
The hardware of the video facility is about to be extended with an extra Primagraphics processor that incorporates Silicon Graphics Elan graphics boards, hardware JPEG compression, FDDI network access and extensive memory. This will add to the video system the capability of rendering 3D images from scene descriptions, and allow preview of video sequences over high speed networks. These enhancements will be the subject of a future FLAGSHIP article.
For further information on the technical aspects of producing input for the facility, on the availability of the documentation and on how to use the time funded by the UFC, please contact me (e-mail: CDO@UK.AC.RL.IB, phone 0235-44-6565).
Chris Osland RAL CCD Graphics Group
The Trainee Scientist (Computing) Scheme was started in 1981 to provide the Laboratory with suitably trained computer programmers. Trainees are recruited with a minimum of two A-levels on a fixed-term, two-year contract. The first six months of their course is spent on intensive training organised by the Central Computing Department. The basic aim of this training is to introduce the trainees to the work of the Laboratory and to see that they have enough knowledge to enable them to become effective members of staff.
They become familiar with many of the machines, systems and languages used on site. They learn the importance of writing structured, commented code which is well documented and thoroughly tested. Fortran 77 and C are the two main languages taught, together with the VM/CMS, VMS and Unix operating systems. They are also introduced to databases, graphics and libraries. Training is given by lectured courses and video with many hands-on practical sessions including several mini-projects. Following the initial training period, the trainees undertake three six-month computing projects, usually in different Departments within the Laboratory.
During the two year course, the trainees are expected to study for a BTEC Higher National Certificate in Computer Studies, attending one of the local technical colleges on one day a week. This provides the necessary qualification for them to apply for a permanent Scientific Officer post.
Over 70 trainees, including those currently on the course, have been through the scheme since it started. Since 1987 the scheme has been run by Garry Williams, who retired at the end of December. Many of the past and present trainees, including the very first two, gathered at the Atlas Centre on December 4th to say farewell to Garry and to wish him a long and happy retirement. A framed copy of the group photograph, shown above, was presented to Garry as a lasting memento of his association with the Trainee Scheme of which he made such a success.
Brian Davies gives photograph to Gary Williams
Tim Pett, Head of Marketing Services Group
The SuperJANET articles that follow are re-printed from the SuperJANET Special Edition of Network News, with permission from the Joint Network Team.
UFC Press Release
Universities Funding Council Gives Go Ahead for Advanced Communications Network
The Universities Funding Council (UFC) announced today (10 November) the development of SuperJANET, an advanced communications network for higher education in Great Britain. The contract for the new venture which has been won by BT is worth Â£18 million over four years, subject to a review at the end of the installation of the pilot phase. SuperJANET will offer higher education and research institutions in Great Britain the opportunity to access the most advanced communications service to support learning and research activities.
SuperJANET, which uses high performance optical fibre technology can be used to transmit voice, data and images. The applications available initially will include: high performance computing, distance learning, electronic publishing, library document distribution, medical imaging and multimedia information services. A pilot network will be established in January to March 1993 which will connect six sites: Cambridge, Edinburgh and Manchester Universities, Imperial College of Science, Technology and Medicine and University College, London, and the SERC Rutherford Appleton Laboratory. Additional sites will be added during 1993-94 reaching a total of around 50. Further sites may be added subject to additional funding.
SuperJANET will complement the Joint Academic Network (JANET) which has served the academic community since 1983 on 200 sites. The advantages of the new network lie in its speed and the types of applications that it can support. SuperJANET can transmit up to 1,000 million bits of information per second, about 1,000 times faster than the JANET. This high speed allows the transmission of highly sophisticated image and voice communications in addition to data.
BT and the UK academic community will collaborate to develop an advanced broadband switching platform for SuperJANET based on ATM (Asynchronous Transfer Mode). BT will provide an optical fibre network together with a range of services to support the development of SuperJANET. The aim is to maximise the number of sites connected and offer a range of access speeds to meet different site requirements. The academic community will help BT to test and pilot new broadband services.
The contract with BT covers the provision of an SDH (Synchronous Digital Hierarchy) network serving up to 16 sites and offering access at 155 Mbps and 2 x 155 Mbps, with trunk network performance of up to 622 Mbps. Prior to the availability of the SDH network the sites will be provided with 140 Mbps PDH (Plesiochronous Digital Hierarchy) capacity. The SDH network will be complemented by BT's SMDS (Switched Multimegabit Data Service) network offering access initially at 10 Mbps. The total number of sites connected is expected to exceed 45.
The following rollout schedule is planned:
- March 1993
- Provision of 140 Mbps PDH network to six sites to create a Pilot Network. The sites that will be connected are Cambridge University, Edinburgh University, Imperial College, Manchester University, Rutherford Appleton Laboratory and University College London. The academic community's plans for the Pilot Network include a high performance data network and a pilot ATM network.
- September 1993
- Extension of the PDH network to serve an additional six sites, namely the Universities of Birmingham, Cardiff, Glasgow, Leeds, Newcastle and Nottingham. A further four sites may be selected for connection to the PDH network at this stage. In addition the PDH sites will be connected to the pilot SMDS network.
- November 1993
- Provision of the pilot SMDS service to a minimum of 29 additional sites. Sites to be selected.
- December 1993 onwards
- Phased introduction of SDH to the sites connected to the PDH network.
- Early 1994
- Introduction of the full SMDS service to all sites.
The ATM work is expected to start with pilot activities forming part of the March 1993 Pilot Network. The collaboration with BT will include the study of how the SMDS service could migrate to an ATM platform and the study of how such wide-area ATM services will interwork with local ATM networks. The rollout plans for the ATM network will be dependent on the outcome of the studies and pilot activities.
BT also plans to include SuperJANET within its recently announced European ATM trial with the Public Network Operators in France, Germany, Italy and Spain during 1994.
Bob Cooper, Director of Networking, Joint Network Team and Chris Fendick, Marketing Manager for High Speed Services, BT
SuperJANET Pilot Applications
The first milestone in the development of SuperJANET will be the provision of a pilot network in March 1993. This is aimed at stimulating and supporting a wide range of new applications that can exploit the high performance, multi-service capabilities of SuperJANET.
The ISC has selected six sites to participate in the pilot phase: Cambridge University, Edinburgh University, Imperial College, Manchester University, Rutherford Appleton Laboratory and University College London. A meeting of these sites was held in July to identify possible application areas that could exploit the pilot network, a second meeting in October discussed a range of project proposals.
In addition to supporting the specific applications, the pilot network will also aim to provide a high performance infrastructure for general use by the six sites and, indirectly, for a wider community. The pilot network will interconnect existing site facilities, such as site FDDI networks, and interconnection with JANET will also be provided. The pilot should be capable of supporting, for example, general use of inter-site X-Windows and some inter-site video communication. The video capability could be used to promote a range of collaborative activities such as shared lectures and seminars, video-conferencing, etc.
The current range of pilot applications is summarised below.
It is expected that SuperJANET will support a wide range of teaching and learning activities. University College London has unique facilities to pilot the use of SuperJANET for teaching surgery via interactive video. It is proposed to set up a series of surgical demonstrations between University College London and other centres on SuperJANET which could support a range of interactive activities including surgical demonstrations from the operating theatre, clinical demonstrations in the lecture theatre, discussions between surgical specialists, the cooperative preparation of video teaching material and exploration of ways of using existing courseware designed for self-directed learning in a distance learning environment. The teaching of surgery, and most medical topics, is heavily dependent on transferring visual information, both static and moving. The source of patients that can be used in teaching is limited and expensive, and changing patterns in hospital training are reducing the variety of patients available to teaching centres. The use of interactive video and the sharing of facilities will in future be required to provide students with the required range of information. The proposal has the support of the departments of surgery at St Mary's Hospital Medical School (Imperial College) and the universities of Edinburgh, Glasgow, Manchester and Cambridge.
Supercomputer Data Visualization and Interaction
Supercomputers are capable of generating enormous amounts of data and in many applications the data can only be understood by the user if it is presented as an image, often using colour and animation. The following are examples of supercomputer applications in which the performance of SuperJANET can be used to support advanced data visualization for remote users.
Display on a remote workstation the real-time output of quantum mechanical computations running on the Cray Y-MP at Rutherford Appleton Laboratory. The displays will include bending patterns of molecules, vibrational modes and electron density maps. The pilot will help to determine what kinds of visualization are feasible over a wide-area network, a successful outcome could result in several other university-based computational chemistry groups wishing to use similar techniques.
Oil Reservoir Studies
The viscous fingering technique can be used to study the recovery process that may be applied in North Sea oil reservoirs. The application will demonstrate how SuperJANET performance could be exploited to provide a quality remote visualization package for the interactive modelling and investigation of viscous fingering. This could be of significant interest to the oil industry. The application will run on the Connection Machine supercomputer at Edinburgh University.
Computational fluid Dynamics (CFD)
The application will demonstrate the use of advanced visualization techniques in the prediction of flow and combustion in reciprocating engines. Large scale CFD calculations will be performed on the Cray Y-MP at Rutherford Appleton Laboratory with visualization output transmitted to Imperial College over SuperJANET.
Global Atmospheric Modelling
Several universities are collaborating in the development of a computer model of the earth's atmosphere that will be used for climate research. The software runs on the Cray Y-MP supercomputer at Rutherford Appleton Laboratory and the SuperJANET pilot will involve users at Cambridge University. The pilot will demonstrate the rapid viewing of a time series of weather maps generated by the Cray. It will provide the remote user with the ability select straight line segments on a map and quickly obtain a display of certain variables in the corresponding vertical plane. A successful pilot is expected to result in requests to use the visualization facilities at other sites.
Library Document Distribution
The cost of academic journals is rising well above normal inflation leading to subscription cancellations and the failure to provide new titles required for research. The cost of storing paper volumes on a long term basis is also causing serious concern in many university libraries. Resource sharing between institutions is seen as one possible long-term solution but existing inter-library loan arrangements are slow and labour intensive. Document delivery over SuperJANET offers the possibility of a fast and cost-effective service. The project will aim to demonstrate document request and delivery over the network involving up to seven universities (including two former polytechnics), with a view to establishing a regular service between all UK university sites as SuperJANET connections become available. The project could be extended to cover other forms of document, e.g. theses or electronic graphics. The potential for long-term cost savings in library shelf-space is considerable.
The aim of this project is to demonstrate the potential of remote access to special documents for real-time consultation between experts at different locations. An important aspect is the extremely high image resolution required to accurately reflect the contents of the documents. The types of documents used in the initial pilot will include Genizah fragments and illustrated Persian manuscripts held at Manchester University. The British Library has expressed an interest in contributing to the project by supplying very high quality images of some rare items in its collections. An advantage of the technique is to allow convenient and widespread access to rare or precious documents without the risk of damaging the documents.
Electronic Journal Testbed
It is proposed to create an experimental electronic journal testbed on SuperJANET through the collaboration of a number of publishers, including learned societies, university publishers and commercial publishers. Each participating publisher will contribute a body of current journal material in a subject specialism. The journal will have a broad subject base and will have features varying with subject specialism in order to explore fully, and take advantage of, the facilities that SuperJANET provides. The range of issues covered by the project could include, for example, the use in science research articles of mathematics, chemical formulae/structure, graphics, half-tones and micrographs. In the humanities the project could cover the use of colour, musical notation, maps, footnotes, etc.
Pathology Consultation Network
The high cost of consultant pathologists makes it difficult, if not impossible, for a health region to employ sufficient pathologists to cover specialised areas; there are, for example, only two or three consultants in the UK covering the pathology of the bone. The rarer species of pathologists are often found in universities, and SuperJANET could therefore be an enabling facility to allow university pathologists to provide a consultancy service in rare pathologies for the whole country. The pilot project aims to demonstrate the effectiveness of the concept by linking pathologists to operating theatres via multi-media workstations coupled with microscopes. The transmission of high quality microscope images and audio/video communication between the operating theatre and the remote pathologist will be via SuperJANET. The project is led by pathologists at Manchester University.
Access to Remote Facilities
This area is aimed at demonstrating remote access to shared facilities. Techniques are now available which can acquire images of the human brain revealing structure and function in exquisite detail. The volumes of data involved are very large and will become larger; new developments will soon permit, for example, the acquisition of whole brain images every few milliseconds. The MRC Unit at Hammersmith Hospital has several advanced facilities for producing brain images that are of interest to research workers at other sites. In the pilot project, brain images acquired at Hammersmith will be transferred via SuperJANET to the Departments of Psychiatry at Edinburgh University and University College London for analysis and to the supercomputer centre at Edinburgh University for processing.
Remote Sensing Data
This project would demonstrate the use of SuperJANET to support inter-site transmission of images of the earth's surface generated by the high performance imaging radiometer developed by SERC and flying on board the ESA ERS-1 satellite. The images are held in a computer database at Rutherford Appleton Laboratory. The availability of SuperJANET will allow remote users to search and browse the large image files, a style of access that is currently inhibited by inadequate network performance. This requirement is particularly important for the Geophysical Data Facility at Rutherford Appleton Laboratory which is required to provide satellite imaging data to a number of universities. The user sites involved in the pilot will include University College London and Imperial College, where the classification of cloud images is one particular area of study associated with research into global warming. Other high-quality and high volume geophysical datasets will soon be available and the interactive dissemination of this data to a dispersed user community is a very important requirement that SuperJANET can support.
This application will aim to demonstrate the potential of a set of networked advanced workstations to support group communication from the users' desktops. The Pandora workstation, developed by the Olivetti Research Laboratory in Cambridge, is a state-of-art multi-media workstation capable of support video, audio and data communication. It is used within the Olivetti Laboratory and Cambridge University. SuperJANET will enable the workstations to be used between widely dispersed sites. The project will also investigate the potential of interconnecting multi-media workstations to studio type audio-visual facilities used for video conferencing and distance learning. It is proposed to use SuperJANET to interconnect the Pandora network at Cambridge to the University of London video network via University College London.
Demonstration of collaborative research and teaching
One of the grand challenges of chemistry is understanding the three dimensional structures of complex molecules and how they interact with smaller molecules such as drugs and metabolites. Recent advances in workstation development, coupled with the availability of SuperJANET, provides the opportunity for sites to collaborate in providing new facilities in this area. The project will use the computerised multi-media facilities developed at Imperial College to create a distributed multi-site facility to support collaborative research and teaching in molecular chemistry.