Date of Award

2009

Publication Type

Master Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

Applied sciences

Supervisor

Angela Sodan

Rights

info:eu-repo/semantics/openAccess

Abstract

In recent years, a significant amount of research has been done on job scheduling in high performance computing area. Parallel jobs have different running time and require a different number of processors, thus jobs need to be scheduled and packed to improve system utilization. Scojo-PECT is a job scheduler which provides service guarantees by using coarse-grain time sharing. However, Scojo-PECT does not provide process migration. We extend the Scojo-PECT by migrating parallel jobs based on system-level checkpointing. We investigate different cases in the Scojo-PECT scheduling algorithm where migration based on system-level checkpointing can be used to improve resource utilization and reduce job response time. Our experimental results show reduction of relative response times on medium jobs over the results of the original Scojo-PECT scheduler and the long jobs do not suffer any disadvantage.

Share

COinS