Experiences in Running Workloads over Grid3

Running workloads in a grid environment is often a challenging problem due the scale of the environment, and to the resource partitioning based on various sharing strategies. A resource may be taken down during a job execution, be improperly setup or just fail job execution. Such elements have to be taken in account whenever targeting a grid environment for execution. In this paper we explore these issues on a real grid, Grid3, by means of a specific workload, the BLAST workload, and a specific scheduling framework, GRUBER an architecture and toolkit for resource usage service level agreement (SLA) specification and enforcement. The paper provides extensive experimental results. We address in high detail the performance of different site selection strategies of GRUBER and the overall performance in scheduling workloads in Grid3 with workload sizes ranging from 10 to 10,000 jobs.