A resource-oriented grid meta-scheduler based on agents

A computational Grid, as an emerging framework for providing globally distributed computing resources to our desktop, has been established as a new programming environment. In particular, a new Grid research for utilizing a large number of busy computers is indispensable for further development of Grid. In this paper, we concentrate on Grid scheduling problem. To cope with unstable availability, low reliability, and mixed operation policies that are still serious problems in metascheduling, we propose ASF, Agent-based Scheduling Framework whose idea is dual to the conventional metascheduling. ASF is composed of a single discreet metascheduler and a collection of autonomous agents attached to each computing resource manager. Each agent autonomously finds jobs to be processed, instead of being assigned a job to by a omnipotent metascheduler. We implement a prototype of ASF, and prove its effectiveness. Our experiment shows that the total elapsed time of job processing is reduced by 11%.