A knowledge-based approach to scheduling jobs in metacomputer environment

Abstract

The importance of metacomputing is without question. Only such a use of distributed resources, as in metacomputing, has the potential to maximize performance and cost effectiveness of a wide range of scientific and distributed applications. There are several projects that address the metacomputing area. Globus, Legion and HPCM are only a few examples. However, in most cases resource dynamics are not taken into account. In this paper we present a concept of the metacomputer management software based on the expert system techniques. The knowledge about some specific aspects of metacomputing, like system specific knowledge, communication networks parameters, architecture specialization, run-time scheduling techniques, optimal resource allocation etc. is kept and managed by different experts. They negotiate with each other to make up a common scheduling decision.