Scheduling in Shared-Memory Multiprocessors

Abstract

Starting with this chapter, we look at how the hierarchical organization can be used to derive efficient job scheduling policies for parallel and cluster systems. In this chapter we describe the hierarchical scheduling policy for shared-memory systems. We compare its performance with a space-sharing policy called Equipartition and a variation of RRJob time-sharing policy. We show that the hierarchical scheduling policy performs considerably better than Equipartition and RRJob policies over a wide range of system and workload parameters of interest. We show that this is true even in the presence of synchronization locks.