A List Scheduling Algorithm for Scheduling Multi-user Jobs on Clusters
Jorge Barbosa (Universidade do Porto, Faculdade de Engenharia, Departamento de Engenharia Informática)
António Monteiro (FEUP-DEI, Instituto de Engenharia Biomédica)
This paper addresses the problem of scheduling multi-user jobs on clusters, both homogeneous and heterogeneous. A user job is composed by a set of dependent tasks and it is described by a direct acyclic graph (DAG). The aim is to maximize the resource usage by allowing a floating mapping of processors to a given job, instead of the common mapping approach that assigns a fixed set of processors to a user for a period of time. The simulation results show a better cluster usage. The scheduling algorithm minimizes the total length of the schedule (makespan) of a given set of parallel jobs, whose priorities are represented in a DAG. The algorithm is presented as producing static schedules although it can be adapted to a dynamic behavior as discussed in the paper.
Parallel and Distributed Computing, Cluster Computing
Toulouse | France | 2008 | June | 24  25  26  27