Hybrid Preemptive Scheduling of MPI Applications on the Grids

— Time sharing between all the users of a Grid is a major issue in cluster and Grid integration. Classical Grid architecture involves a higher level scheduler which submits non overlapping jobs to the independant batch schedulers of each cluster. The sequentiality induced by classical batch scheduling is not fitted for the number of users and the heterogeneity of the jobs in Grids. Time sharing techniques address this issue by allowing the simultaneous execution of many applications on the same resources. Classical sharing techniques are co-scheduling and gangscheduling. Co-scheduling rely on the independant operating system of each node to schedule every process of every application. Gang-scheduling ensures that the same application is scheduled on all nodes simultaneously. Previous work has proven that co-scheduling techniques outperforms gang-scheduling while physical memory is not exhausted. In this paper, we introduce a new hybrid sharing technique providing checkpoint based ex...

• Type: Conference
• Year: 2004
• Where: GRID
• Authors: Aurelien Bouteiller, Hinde-Lilia Bouziane, Thomas Hérault, Pierre Lemarinier, Franck Cappello
• Keywords: GRID 2004, Physical Memory, Sharing Technique, Time Sharing