Module 3: Massively Parallel Systems

Areas of excellence

Software for parallel computers
Special-purpose architectures, new computer principles

Projects

Description

The technological limits of computers with just a few very fast processors are not far off. At the same time, the demand for higher computing power in some major areas of application has risen sharply. This challenge is met by known parallel-computing architectures. Though parallel systems are already available as far as hardware is concerned, simple and efficient programming to take efficient advantage of the large architecture potential in powerful applications remains problematic. At present, the path to massively parallel systems is augmented by architectural and programming language ideas which aim at a basic departure from familiar operational principles. In this light, the research projects in Priority Programme Module 3 can be divided into two large areas of excellence:

Software for parallel computers
Special-purpose architectures, new computer principles.

Software for parallel computers

How to improve software for multipurpose architectures? Quality attributes requiring attention in this area are: correctness, efficiency, portability and development effort. But efficient algorithms and the pooling of them in libraries is just as important. Consequently, Module 3 also includes projects which put algorithm questions and application areas in the foreground. This field features the following three project groups.

The Theory/Methodology group is roughly concerned with the development of practicable formal methods for designing correct parallel and distributed algorithms and demonstrating their applicability and use in connection with practical problems. Other projects seek to find more efficient solutions to very difficult problems (NP-hard problems) working from probabilistic angles or axiomatic specifications.

In the Methodology/Tools group, one project is devoted to the development, implementation and evaluation of a co-ordination language capable of describing the process and synchronization properties of parallel algorithms. In another case, such questions are addressed with collections of reusable "skeletons". Here the aim is to provide the programmer with the most important basic classes of parallel algorithms, in the form of a classification, which he completes as necessary for the finished application.

The Algorithms and Applications group features such projects as the development of algorithms for solving linear equation systems by divide-and conquer methods and the production of a compute-server to boost the performance of mathematical packages in workstations. Finally, one project is researching links between computational chemistry and computational geometry and establishing methods for configuring molecules in given conditions. The complexity of the subject-matter requires the use of novel parallel algorithms.

Special-purpose architectures/new computer principles

Special-purpose architectures are designed for specific, defined applications, with development limited to those tools actually needed for the particular job. This avoids a lot of unnecessary ballast, with the hope of achieving the desired goal more quickly and efficiently. Besides employing specialized processors, this area of excellence also applies completely new computer principles.

Multiprocessor/Multidisc System designates the development of an image server, embedded in the computer network, which takes care of high-resolution image management. With hypermedia environments gaining in importance all the time, this project has high practical implementation potential and already enjoys industrial backing.

Signal processors provide the basis for pushing ahead with the development of a high-performance computer suitable for special tasks, at a moderate price, thanks to the technology involved. The hardware has already been developed in earlier work and has shown impressive results, while its programmability is being demonstrated in conjunction with other projects.

Artificial neural networks have received little research attention in Switzerland to date. The Priority Programme supports a centre of excellence marked by co-operation among theoreticians, hardware designers, software developers and others through to users. Impressively, concrete applications are already crystallizing. Electricity distribution network operators are among those who are pinning their hopes on neural networks.

Interdisciplinary answers can be expected in addition to the results achieved in the individual research projects of Module 3. It will be interesting to see how much interaction there is between the various areas, for example between mainly basic research-oriented projects and those concerned with converting their finding into programming tools. In the past, research groups in Switzerland have worked in isolation. The aim for the next two years is to foster a much more closely knit research community in the field of parallel systems.

[Home]

[List of Projects]