HPCN project manager Patrick Van Hove said the challenge for the HPCN industry today is to enlarge the potential of technology by bringing its potential benefits to all industrial users.

Van Hove said: 'Barriers to the wider use of high performance computing are mainly due to lack of awareness, lack of success stories, and the perceived complexity of HPCN.'

Van Hove said the commission is determined to build on the success of the Europort project in the Esprit R&D programme which has demonstrated the use of HPCN in real life applications.

'A large number of users selected industrial applications and with the help of some of the best expertise in Europe they tuned and tested these applications on parallel machines,' he said.

'In many cases they demonstrated that with say 10 workstations working in parallel and with clever software, a substantial speed-up was achieved,' he said.

Van Hove said the commission was about to start a number of actions to promote the results of Europort and other HPCN projects.

He said efforts would be made to evaluate the feasibility of new applications and to transfer the successes from one domain to another.

The development of user-friendly tools to enable non-specialists to tune and de-bug HPCN sytems should be a major target for the latest tranche of Esprit projects and elsewhere, Van Hove said.

'These tools exist in research but they must be consolidated, brought to the market and used in industry,' he said.

Peter Chapman

Thinking Machines vice-president for marketing Peter Masucci gave the lead presentation on 'Applying Parallel Technology to the Real World'. It emphasised the way that the restructured supercomputer maker is marketing workstation-based clusters for business users. Masucci explained how his company has developed a software package for data mining.

It can work in parallel processing systems and handles terabytes of data with records containing thousands of fields. It uses several methods to analyse and classify data.

For example it uses neural networks for classification and high order correlations, genetic algorithms for model generation through optimisation, and the StarView product to visualise data and view the results.

Another vendor presentation was given by Eric Theron from France's Matra Cap. This presentation described CapCASE, a graphical development environment for high performance real time applications. The idea is that you can use this package for visual heterogeneous source generation.

CapCASE claims instant portability without loss of efficiency. It uses software and hardware graphs to enable it to produce code generation. Matra says it protects previous user investment by minimising software porting costs to other systems.

The package has been developed over the last 10 years with money from three separate projects. The first two were EU-funded via the Esprit progamme. These are ARVISA and CLEOPATRA. The third is the EUROTOPS project, supported by the pan-European R&D EUREKA programme.

The CapCASE product has been commercialised and licenced to several major computer companies.

Chris Lazou

The Applied Parallel Research products covered by the distribution agreement are FORGExplorer interactive tools, xHPF, which is a high performance Fortran compilation system, and spf, a shared memory parallelisation system.

In addition, Visual Numerics is to make available its the IMSL Fortran 90 and Fortran 77 numerical Libraries to run on the IBM SP/2 parallel system. The two Visual Numerics products will also be available to run on IRIX 6, the operating system of the Power Challenge of Silicon Graphics.

For the uninitiated, data warehousing is a term for the storage of large quantities of data. Data mining is using this data to extract intelligent information for management decisions. Ideally these decisions are in real time. For example, banks like to have customer data at both a central 'warehouse' and at the local branch for the manager.

Business managers want to have client profiles so they can serve their clients individual needs. For example, supermarkets can easily use barcodes to collect information on the purchasing habits of customers who pay by credit cards and whose addresses are known. Then the stores can ensure that customers favourite goodies are available on the right shelf.

The first speaker to emphase the advantages of these techniques was Dr Michael Frick, from Andersen Consulting Germany. He said the biggest benefits achieved from data mining are the enforced change of a business's organisational culture. You need to re-engineer the customer's organisation to make it more sensitive on managing costs and risks.

Then Frick went on to describe the tricks of how to sell the idea to top management so they can authorise the incorporation of data mining experience in the business methodology.

Frans van de Ven from Coopers & Lybrand, Belgium, emphasised the capacity of high performance computer solutions to deliver integrated management information. Data collection involving of the order of 300 gigabytes are needed for applications dealing with individual clients. He went on to describe how to implement such a facility, and how to manage the risks.

The final advice was to carry out this out implementation step-by-step, to manage expectations, and not to try to oversell it.

Dr Georg Bolz from Lufthansa Systems GmbH, Germany, said his airline's problem was how to schedule flights, crews and passengers. The problem can be split into strategic planning, pre-schedule optimisation, International Air Transport Association (IATA) draft schedules, ad hoc changes to accomodate IATA requirements, and the final operational plan.

The whole model is evaluated for profitability and all the optimisations are tested against this criterion. Parallelising the market share model allows the scheduling of a whole journey rather than single operations. He concluded that parallel systems are necessary for a proper interactive response.

Colin Upstill from the Parallel Applications Centre in the UK said the commercial benefits from parallel systems are already with us. He gave examples of brewers, shoe makers and banks which are users of this technology. Some bought systems costing a million dollars and have recouped their costs in a few months. The high return on investment is therefore already demonstrated.

There are still many drawbacks to using this technology, such as the lack of professionals who understand how to exploit high performance computers. There is also the difficulty of establishing credible expectations, stripped of 'hype'.

He concluded that 'the benefits achieved are considerable and the biggest risk for businesses is to ignore this technology'.

Chris Lazou

In this not-too-short article one could read that a T-flop machine will be installed in the Palais des Congres and that the machine will run the test. Furthermore the article muses on the use of supercomputers and explains what 'Tera' means and so on. Yours truly is also featured as an expert explaining that supercomputers can be used for non-belligerent subjects.

What happened? Bad journalism mainly. The core of the article is similar to one I wrote on the last HPCN Event in Milan. As most of the readers will probably remember, Ed Masi from Intel was one of the keynote speakers. Masi drew attention to a programme of a U.S. government agency about which very few people in the audience had heard. That program, called ASCI (Accelerated Strategic Computing Initiative) is designed to fuel the next wave of high-performance computing investments in the USA.

Put very simply, the objective of ASCI is to provide a substitute for underground nuclear weapons testing in order to ensure the safety - as far as bombs can be called safe - and preparedness of these strategic assets. Intel has won the contract in which funding will start with 35 mln ecu jumping quickly to 75 and 100 mln ecu.

The system will deliver ten times the performance of today's fastest supercomputers. More than 9,000 of Intel's next-generation microprocessors - in September still code-named P6, now officially called the Pentium Pro - will power the system, to be located at Sandia National Laboratories in Albuquerque in the state New Mexico.

The Morgen article followed this with some explanations and theory on high-performance computing and networking plus an encouragement to go to Brussels in April, to meet the expertissimos, and to hear the discussions.

This all read sloppily and was put together hastily - they even spelt my name wrong - and the result was laughter among HPCN participants able to read Flemish plus a threat by an anti-nuclear group (who are against bombs, simulated or not) to pay an unwelcome visit to the Palais des Congres. They did show up and talked to the conference event chairman, Bob Hertzberger. Apparently he succeeded to talk some sense into them and they left.

Jaap Hollenberg

NetStar's GigaRouter product helps the information in the Internet travel more efficiently. It relies on the traditional small packets of data, each with an IP address. The job of a router is to look at this address and ensure the data arrives at the correct destination.

The GigaRouter is more than ten times faster than others and copes with different types of net connections such as HIPPI, FDDI, ATM and HSSI. With the increment of people connected to the net it is impossible for little routers to have in its tables all the addresses that are in the Internet.

'With the GigaRouter we can have all the addresses in the world in the machine', explains Steve Markham, general manager in Europe of NetStar International Ltd.

The uses for the router are essentially with three types of customers: telephone companies and Internet service providers; supercomputing centers; and companies or organizations that have hundreds of worstations and want to connect them together.

'We have now about a hundred GigaRouters sold to these kind of customers', Markham said.

NetStar is also responsible for the first ATM (asynchronous transfer mode) connection from the east to the west of the U.S. which is called the very high bandwidth networking service. Next, NetStar, is working on an ATM OC-12c card which is four times faster than the other ATM connection.

M.Àngels Novoa

'Vampir is easy to use because is very graphic', said Werner Krotz-Vogel, senior analyst of Pallas. Anyone using the software can see how the communications changes when parallel software is running and the details are provided with a zoom just by clicking on a certain part of the graphic.

Vampir makes it possible to simulate benchmark testing of different software on a workstation instead of having to use a bigger computer. 'With this it is possible to know - with an 80-90% degree of certainty - the performance of a particular software on different machines without actually working with them', said Krotz-Vogel.

Pallas is demonstrating the latest MPI message passing interface, already taken up a wide community of scientific and industrial experts. Pallas is offering MPI implementations, software development tools and professional end-user support at the HPCN exhibition. M.Àngels Novoa

'The system is very complete, it includes a SPARC chip front-end and all the accessories such as keyboard, mouse, hard disc and ethernet connections. This is a supercomputer at a desktop size,' explains Carlo Alberto Marchi, marketing and sales manager in Europe for Alenia Spazio.

Alenia offers the latest machine as part of its Quadrics line of systems. It is a line of SIMD architecture processors that are scalable from 8 to 2.000 computer nodes. They have a peak performance between 0.4 to 100 Gflop/s.

'We have installed more than 50 systems just in Italy and Germany since 1993 and recently we have installed our first system in the United Kingdom', says Marchi.

The latest machine from Alenia joins its Quadrics/MC machine, developed for specific aplications and introduced in the market a year ago.

M.Àngels Novoa

Trapper is a graphical programming environment for parallel systems. It offers a graphical design tool and a configuration tool that produces the configuration files needed by PVM. PVM is a visualization tool to gauge the behaviour of the application as a performance tool.

'It is difficult to begin to write parallel software and we want Trapper to make it easier', said Johannes Grawe, from Genias Software. For more information contact Elke Thur at the e-mail address thur@genias.de or check the information in the web page http://www.genias.de. M.Àngels Novoa

NLR's Loeve flies on as Dutch aerospace labs hit by Fokker crash

Loeve says the NLR can survive the aftermath of the recent bankruptcy of airline manufacturer Fokker - its biggest customer. But he admits jobs will be lost when the shock waves reach NLR.

That might not be until 1998, Loeve said, because Fokker is committed to build the planes it has already agreed with airlines.

'The strange thing is that demand for Fokker aircraft has increased in the last few months. People see that they are very beautiful aircraft and they want to buy them while they still can.

'It is true that jobs will have to be lost. In 1998 we are projecting that we will have 150 less people than at the moment. We currently have 950 staff and that will go down to 800,' he said.

Loeve admitted that even his best qualified staff would not be immune from the cuts.

'The problem is that we have a lot of mono-disciplinary staff - experts in one particular field. If there is no longer a need for their skills after Fokker then these people will lose their jobs. ' he said.

The Fokker business accounted for a third of NLR's 150 mln guilder ($100 mln U.S.) sales. Loeve thinks he knows why his customer went bankrupt.

'They went bust because of the cost of aircraft design is so much. Building aircraft does not cause a loss,' he said.

NLR supplied Fokker with expensive software for aircraft design such as computational fluid dynamic tools as well as wind tunnels, and flight test equipment for the certification of aircraft.

But with a hole in its aircraft design business where will NLR plug the gap left by the loss of Fokker business? Loeve believes there's enough other work to go round.

'We are the Dutch lab for the support of the aircraft industry and support of agencies in traffic management. We work for the military and airline companies such as KLM and Sabena,' he said.

'We do fluid dynamics simulations, flight simulations, and help the military prepare their missions. The kinds of software we have developed enables the military to plan their fuel consumptionand to plot their way through so-called ''standard way points" that planes travel along,' he said.

'We also produce the systems for electronic counter measures such as the chaff that aircraft throw out to put missiles and radar off the scent,' he said.

Another area where NLR can win business is in the satellite sector where NLR develops flight critical software.

'The risk is very high. In satellites there is a lot of money involved and you can't afford for things to go wrong,' he said.

Loeve has been in supercomputing at NLR since it bought its first NEC SX-2 machine in 1987.

He has some contentious views about the role of high-performance computer technology. Loeve says he prefers vector processing to parallel processing despite the trend towards work stations and parallel.

'I have been telling people for a long time that the cost to performance ratio is better with vector processing than for workstations. This view is contrary to the view at the HPCN show,' he said.

Loeve does not believe the days of the big expensive machine are numbered. The latest 32-bit processor NEC SX-4 that Loeve and his team has ordered for July this year has a list price of U.S. $20 mln. But Loeve argues that it is well worth it if you have the workload to justify the expense.

'Seventy percent of our contracts that involve the use of our supercomputers - that is where we earn our money. It makes sense to have a really fast machine if you have enough customers.

'Also it is less work for our people who have to deal with the system,' he said. Loeve is keen to spread the word about his vision of the integrated system. He is critical about what he sees as a 'bottom up approach' to the use of equipment. 'What you have is a lot of users autonomously organised with a large say in choice of system. Locally, where there is a lack of money they choose the best they can,' he said. Loeve believes top management and government organisations should take the lead and install high performance equipment. Costly to buy - but when it is spread across a whole organisation, it becomes a value-for-money option, he said.

'Management is to blame. But also there are vendors concentrating on boxes and not systems integration aspects,' he said.

Loeve advocates a single system approach with only one 'virtual computer' linked to terminals replacing the large number of servers spread across a network.

Peter Chapman

How to subscribe to Primeur Live! On the Web, Primeur is accessible at the address: http://www.hoise.com/primeur You can also check in at: http://www.hoise.com to see which information on HPCN in Europe is available. The Web-server will be updated continuously, so you can check in whenever you like to see what's happening in Brussels. Each day we will produce a newsletter with the most important news headlines and send it out through the e-mail. Two versions of Primeur Live!! are available: Primeur Live! html Primeur Live! plain You can read the HTML version with your favourite WWW-browser (for instance Netscape or Mosaic). The hyperlinks will guide you to more background information. This is a rich source of information. You can read the HTML version even if you do not have access to the real web: you only need the browser on your machine. The PLAIN version is for people who do not have a WWW-browser. It does contain the headlines, but no links to background information. To subscribe please send an e-mail to the address: emmen@genias.nl with your name and affiliation (name, address, company, telephone, etc.) and the e-mail address where you want to receive your copy of Primeur Live! Indicate whether you want to receive the HTML or PLAIN version.

The team of Primeur Live! Editors Ad Emmen M. Àngels Novoa Nigel Tutt Peter Chapman Your correspondents from Brussels Jean-Loic Delhaye Chris Lazou Uwe Harms Jaap Hollenberg Giacomo Polosa Business Development Terry Davidson Graphic Design Rob Hulsbosch WWW support Elly Molendijk Rick Jansen Copyright The HOISe Consortium