Impact of Fair Share and its Configurations on Parallel Job Scheduling Algorithms

To provide a better understanding of fair share policies supported by current production schedulers and their impact on scheduling performance, A relative fair share policy supported in four well-known production job schedulers is evaluated in this study. The experimental results show that fair share indeed reduces heavy-demand users from dominating the system resources. However, the detailed per-user performance analysis show that some types of users may suffer unfairness under fair share, possibly due to priority mechanisms used by the current production schedulers. These users typically are not heavy-demands users but they have mixture of jobs that do not spread out.

• Sangsuree Vasupongayya

• Fair share
• Parallel job scheduler
• Backfill
• Measures

[1] J. KAY and P. LAUDER, "A fair share scheduler," Communications of
the ACM, vol. 31, no. 3, pp. 44-55, 1988.
[2] S. Kleban and S. Clearwater, "Fair share on high performance computing
system: What does fair really mean?" in Int-l Symp. on Cluter Computing
and the Grid (CCGRID), 2003.
[3] OpenPBS. (Online). Available: http://www.nas.nasa.gov/Software/PBS/
[4] PBS pro. (Online). Available: http://www.pbspro.com
[5] LSF fairshare documentation. (Online). Available: http://accl.grc.
nasa.gov/job schedulers/lsf/Docs/lsf6.1/lsf6.1 admin/E fairshare.html
[6] S. Kannan, M. Roberts, P. Mayes, D. Brelsford, and J. Skovira, "Workload
management with loadleveler," IBM Redbook, Tech. Rep., 2001.
[7] Maui scheduler. (Online). Available: http://www.supercluster.org/maui/
[8] MOAB scheduler. (Online). Available: http://www.clusterresources.com/
products/mwm/docs/MoabAdminGuide450.pdf
[9] S.-H. Chiang and S. Vasupongayya, "Design and potential performance
of goal-oriented job scheduling policies for parallel computer workloads,"
IEEE Trans. Parallel and Distributed Systems, 2008.