Scheduling of MPI applications: Self-co-scheduling

Abstract

Scheduling parallel jobs has been an active investigation area. The scheduler has to deal with heterogeneous workloads and try to obtain throughputs and response times such that ensures good performance. We propose a Dynamic Space-Sharing Scheduling technique, the Self CoScheduling, based on the combination of the best benefits from Static Space Sharing and Co-Scheduling. A job is allocated a processors partition where its number of processes can be greater than the number of processors. As MPI jobs aren't malleable, we make the job contend with itself for the use of processors applying Co-Scheduling. We demonstrate that our Self Co-Scheduling technique has better performance and stability than other Time Sharing Scheduling techniques, especially when working with high communication degree workloads, heavy loaded machines and high multiprogramming level. © Springer-Verlag 004.