SC2000 The state-of-the-field

Sterling remarked that another sign of the growing importance of cluster computing could be found on the SC 2000 exhibition floor where 33 of the 95 commercial exhibitors were in some way involved in cluster computing. Sterling foresaw a swift development to still more mature operating environments and a larger adoption, especially in view of the fast interconnection networks that are available in increasing numbers. In a few years they may equal the bandwidth as used in integrated parallel systems which would make both classes of systems about equal in almost every respect. In Sterling's view the future of clusters was therefore bright and would eventually replace the run-of-the-mill parallel machines but not the few very large special systems built for performance rather than price-performance.

The second talk in this track was had a completely different nature. It was entitled ``A small dose of Infosec'' by the speaker Gene Spafford and addressed the field of computer security, or rather the lack of it. Although entertaining, it was also scaring to see how the inappropriate use of computer systems all over the world is exploding. There are several reasons for this. A first one is the growing number of (interconnected) computer systems {\it per se}. A second one is that the vast majority of computers runs a single OS, some variant of Microsoft Windows. This mono-culture makes a large community vulnerable at once. A third reason is that one does not have to be a real expert anymore to be a hacker. Many of the burglary tools, virus and Trojan Horse codes, etc., are easily accessible from Internet and have even become so ``user-friendly'' that almost anyone can start a hacker career at a fairly advanced level. At this moment ca. 80 viruses/week are released, on average in the USA 50--60 intrusion incidents/day per user occur on Internet (of course very unevenly distributed with NASA and the Pentagon as top attraction points). Spafford made the prediction that by 2004 100,000 would be around of which 99\% for Microsoft and the rate of introduction would be about 1 virus/hour while the number of attacks will have grown to 10/hour. In all, this talk was both entertaining and scaring leaving us with the moral that the average user should become a lot less trusting than we are in the present idyllic days.

Margaret Wright from Bell Labs. was the third person featuring in this series. The title of her talk was ``Numbers, lots of numbers and insight too''. In this title she paraphrased the Tukey's well-known quote ``The purpose of computing is insight, not numbers''. Margaret Wright tied to qualify this quote by showing very interesting and compelling examples of present-day research were numbers and insight are so interdependent that it would be nonsensical to state that the numbers are not relevant. As general examples she mentioned the self-tuning numerical software projects, like FFTW and Atlas, the computation of pseudo spectra, and adaptive optimisation using generalised eigenvalue analysis. She closed with the presentation of an example of a large scale practical project: the optimisation of the coverage of a district with cell phone antennae and the associated sensitivity analysis. This is a very complex problem because it involves geographical data, antenna characteristics, usage and capacity data, etc. In the end indeed an optimal scheme of directing the antennae was found, but it was a perfect illustration that not only insight mattered but that also the numbers were indispensable both because large scale computing was at the heart of the solution but also because of the mass of data to include in the modelling of the problem.

The last speaker in this track was J.C Browne from the University of Austin, Texas. He gave an overview of languages for Parallel and Distributed computing. The number of these languages is bewildering and, in this sense, can be seen as a research success. On the other hand, the adoption of these languages is minimal and therefore an application failure. Browne reviewed some of the language features and interrogated the audience with respect to their usage of for instance MPI, OpenMP, and HPF. Of the certainly large audience, there was between 10--20\% that actually had written programs with these language facilities (for HPF there were only two persons). This was probably not a fair assessment because many people in the audience not (anymore) in the league of program writers while they may be users or decide in having it used. Still, Browne made a point because for the more ``exotic'' languages it was clear that a majority knew them at best by name. The speaker tried to identify some of the reasons for the lack of adoption of languages that potentially have much to offer to parallel programmers in terms of ease of use and/or expressiveness. In most cases the inability to make the end product efficient on a wide range of platforms is the main hurdle. Also, in some cases, a language becomes more or less irrelevant because it was targeted at a specific architecture that has ceased to exist, like Connection Machine-like systems. Browne noticed that new research in the parallel programming language area is stalling because of the awareness that agreed standards, like MPI and OpenMP have been adopted in a large community. although this can be cumbersome and not always efficient. A new language should have very attractive new features to compete with these programming models to have a real impact in the parallel programming community.