advertisement:
 Visit our sponsors

HPCN in the new Millenium -
Special RCI European Management Symposium, December 1998

Primeur Live! is published during major High-Performance Computing and Networking events in Europe.

Primeur is a Virtual Magazine on High Performance Computing and Networking in Europe with monthly and weekly editions. It is produced by an editorial team composed of professionals in publishing and HPCN. You can put the editorial team to work as well. Read about our services and find out about the friendly rates.

Breaking news - just a click away:

HPCN in the new Millennium
Early December 60+ experts on high-performance computing and networking, mainly from Europe gathered in Bristol UK to discuss "HPCN in the new Millennium" at this years RCI European Member Management meeting. The conference being hosted by the British Aerospace Sowerby Research Centre focused on engineering in general and aerospace in particular. Users from main European companies and centres met representatives from nearly all HPC companies currently active. The most important trend noticed, is the revival, or at least the possibility of a revival, for parallel vector machines (PVP) especially for industrial engineering applications that was mentioned by several speakers. Being able to deliver near peak performance on industrial codes that do not parallelize well in general, PVPs are the only systems capable of delivering the necessary performance.
Europe can take the lead again in supercomputing - vectors can be the answer
Dutch scientists now happy working on their Cray C90 did not want to learn parallel processing during the past years. And for good reason. It is difficult, it is time consuming and for their class of problems does not work. They are not the only ones. In industry one finds that a lot of industrial codes just do not parallelize. Good old vectors could provide help here. Japanese vendors like NEC and Fujitsu have good machines out there that can bring a lot of performance with only a few processors. Europe could benefit from that. Since the US have effectively banned Japanese vector computers and do not have a real alternative of their own, Europe, which has long standing experience with these machines, could take of as the leading supercomputing continent. Instead of struggling to get thousands of processors to work as the poor Americans have to do in the Asci project, we could buy Japanese vector machines to do real application work with them. It may not be advanced HPC architectures but it can be used for advanced applications.
In 2004 only Tflop/s machines will make it to the TOP500 list
"Despite its limitations, the Linpack benchmark is still the best yardstick for the TOP500 list" said Prof. Hans Meuer at the RCI conference, "It is simple, available on all systems and is a small real application in itself." According to Meuer, the list managers are open to suggestions and discuss regularly with vendors and other possibilities One problem he sees for instance with the Linpack benchmark is that on vector machines on real life applications often 95% efficiency can be reached as opposed to 20% for MPPs, while they both show high efficiencies on the Linpack benchmark.
HPCN in Europe - A Critical Assessment
There is a crisis in HPCN. During the recent years, the area has become a marginal one in the IT sector. It is no longer the cutting-edge-technology-area that attracts the brightest young people. There is no real progress in parallel computing and HPC systems, supercomputers, are just one option in a large growing catalogue. In Europe, the crisis is even worse: there is no HPC industry, no HPC community, and no European HPC policy. When Europeans active in HPCN meet, it is in the USA. On the positive side, Europe is still a large consumer of HPC technology from the USA and Japan and applications that use a little bit of HPC technology are quit successful.
European aerospace research goes SX-5
Speaking at the RCI symposium, 2nd December 98, in Bristol, UK, John Murphy, head of computational engineering in British Aerospace, said: "The key challenge for the aircraft industry is how to deal with the ever increasing complexity of applications and the way various elements are integrated to produce the final aircraft. New aircraft, be it civil ones, or military, tend to have more complex demands on both material, physics and electromagnetic signatures". With ONERA, France, NLR, The Netherlands and DLR, Germany planning to use NEC SX-5 vector supercomputers, this means that in Europe at least, the Aerospace industry has decided that the NEC SX-5 will satisfy their performance needs without any need for major reprogramming to parallelise codes.
MPP under attack from clusters and PVP
According to Compaq's European HPC marketing manager Martin Walker, clusters have the future and ISV's hold the key. Unix is the preferred system for HPC for years to come. European supercomputer vendor QSW has already installe 0.5 Tflop/s of computing power, and the SX success show viability of PVP machines, said NEC's European marketing manager Chriastian Lantwin.
Data, data, metadata
A few decades ago, all the data that were used to write a scientific article, were included in the article itself. Technical data were available in reference books that summarised all the basic data needed in a specific engineering area. As Kerston Kleese pointed out at the RCI conference, we have come a long way. Specialized meta data centres, as the one at the British Central Laboratory of the Research Councils for which Kleese is responsible, host several specialized data centres.
My Cray C90 or Jedda system needs to be replaced - but by what?
You are running an old parallel vector system (PVP) like the Cray C90 or J90. Your users are happy with the system, but it is becoming too small and too expensive to maintain. Then, according to, RCI advisor Jeffrey Mohr you have a problem. And when you are in the USA, the problem is huge.
Get physical - Distributed HPCN only works when you know your data
Design, however clever and advanced, only has a small effect on the overall cost of for instance a future aeroplane. When the design starts, the constraints given, already account for 70% of the total costs. Pierre Perrier, Dasault Aviation, pointed this out at the RCI Conference. In addition, he showed virtual product design of an aeroplane is very difficult. The reason is that every part and property is connected very closely to every other part and property of an aeroplane. It is difficult to break the design into parts that can be done by separate development groups, which can be at different places.
Supercomputer centres need to redefine themselves
Supercomputer centres get more and more competition from high-performance servers and clusters, which, placed in departments, compete with the centre. On the other hand, really large machines, are expensive and have so much capacity than most supercomputer centres cannot afford them. Colin Upstill, manager of the Southampton Parallel Applications Centre (PAC), sees three ways out: stop, get back to become a large (academic) centre again; or become an IT innovation centre.