Editorial: Portable software tools for parallel architectures
✍ Scribed by Wadsworth, Chris
- Publisher
- John Wiley and Sons
- Year
- 1999
- Tongue
- English
- Weight
- 31 KB
- Volume
- 11
- Category
- Article
- ISSN
- 1040-3108
No coin nor oath required. For personal study only.
✦ Synopsis
Parallel computing is at a crossroads. The hardware exists, the demand for the high performance computing power it offers is all around, yet mapping one onto the other remains an order of magnitude too difficult for widespread takeup. Recent editorials, e.g. O'Leary [1] and Furht [2], have highlighted differing views of the prospects.
Several studies, notably that of Messina & Sterling[3], have identified three main goals for software and tools development for high performance computing (HPC):
-performance -usability -portability Performance is generally seen as the raison d'être of parallel computing, for ever more demanding applications. Usability -or programmability -pertains to the effort involved in developing (parallel) software, the productivity of the developer, and the effectiveness of development tools. Portability protects long-term investment in software development by ensuring common interfaces across different systems and continuity as systems evolve. A further goal, scalability, may be perceived as a special case of portability, i.e. porting to a different number of processors (generally larger, though sometimes smaller).
The ultimate aim is optimum achievement of the three goals in combination: in short, methods and tools for developing portable, high performance software, the methods and tools being effective and productive in practice. Experience to date, however, has shown conflicting design decisions and trade-offs that are not generally resolved.
Portable Software Tools for Parallel Architectures (PSTPA) is the title of a UK programme that has been running for five years (1994)(1995)(1996)(1997)(1998)(1999) supported by the Engineering and Physical Sciences Research Council. While the title emphasises portability, the other two goals are equally essential, and taken as understood. Historically, portability has been given less attention, yet in the end these issues pertain directly to the prospects for a mature parallel software industry, much as sequential software has matured over the last 20-30 years (see O'Leary [1]). For parallel computing, the key question may be stated: (to what extent) can portability be achieved without compromising performance and usability?
The particular aims of the PSTPA programme have been:
-to capitalise on UK strengths in research into the foundations of generic software tools, languages, application generators, and program transformation techniques for parallel computing, -to widen the application programmer base and to improve productivity by an order of magnitude, -to maximise the exploitation potential of tools developed in the programme, -to maximise architectural independence of the tools and to maximise the leverage of open systems standards,