Scheduling Rigid, Evolving Applications on Homogeneous Resources

Cristian Klein (1) , Christian Pérez (1)

1 GRAAL - Algorithms and Scheduling for Distributed Heterogeneous Platforms

Classical applications executed on clusters or grids are either rigid/moldable or workflow-based. However, the increase of resource computing and storage capabilities has leveraged more complex applications. For example, some code coupling applications exhibit changing resource requirements without being a workflow. Executing them on current batch schedulers leads to an inefficient resource usage, as a block of resources has to be reserved for the whole duration of the application. This paper studies the problem of offline scheduling of rigid and evolving applications on homogeneous resources. It proposes several scheduling algorithms and evaluates them based on simulations. Results show that significant makespan and resource usage improvement can be achieved with short scheduling computing time.

Mots clés: Clusters; Resource management; Scheduling; Evolving Application; Rigid Application

Domaines: Calcul parallèle, distribué et partagé [cs.DC]

Cristian Klein, Christian Pérez. Scheduling Rigid, Evolving Applications on Homogeneous Resources. [Research Report] RR-7205, INRIA. 2010. ⟨inria-00456769v2⟩