Memory Conscious Scheduling for Cluster-based NUMA Multiprocessors

Abstract

For Non-Uniform Memory Access (NUMA) multiprocessors, memory access overhead is crucial to system performance. Processor scheduling and page placement schemes, dominant factors of memory access overhead, are closely related. In particular, if the processor scheduling scheme is dynamic space-sharing, it should be considered together with the page placement scheme for efficient process execution. Most research in this area, however, has focused exclusively on either the processor scheduling scheme or the page placement scheme alone without considering the interaction between the two. This paper proposes several policies for cluster-based NUMA multiprocessors that are combinations of a processor scheduling scheme and a page placement scheme and investigates the interaction between them. The simulation results show that policies that cooperate to employ the home-cluster concept achieve the best performance. The paper also compares the best of the proposed policies with other existing dynamic processor scheduling policies. Based on our study reported here, the best policy is found to perform better than other existing policies.

DOI: 10.1023/A:1008165010732

15 Figures and Tables

Showing 1-10 of 21 references

Performance evaluation of two-level processor scheduling in collaboration with memory management for NUMA multiprocessors

  • T. Koita, T. Katayama, K. Saisho, A. Fukuda
  • 1998
1 Excerpt

A survey of scheduling in multiprogrammed parallel systems

  • D. G. Feitelson
  • 1994
1 Excerpt