skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Page placement policies for NUMA multiprocessors

Abstract

In many parallel applications, the size of the program's data exceeds even the very large amount of main memory available on large-scale multiprocessors. Virtual memory, in the sense of a transparent management of the main/secondary memory hierarchy, is a natural solution. The replacement, fetch, and placement policies used in uniprocessor paging systems need to be reexamined in light of the differences in the behavior of parallel computations and in the memory architectures of multiprocessors. In this paper the authors investigate the impact of page placement in nonuniform memory access time (NUMA) shared memory MIMD machines. The authors experimentally evaluate several paging algorithms that incorporate different approaches to the placement issue. Under certain workload assumptions, the results show that placement algorithms that are strongly biased toward local frame allocations but are able to borrow remote frames can reduce the number of page faults over strictly local allocation. The increased cost of memory operations due to the extra remote accesses is more than compensated for by the savings resulting from the reduction in demand fetches, effectively reducing the computation completion time for these programs without having adverse effects on the performance of typical NUMA programs. The authors also discuss some early resultsmore » obtained from an actual kernel implementation of one of the page placement algorithms.« less

Authors:
;  [1]
  1. Dept. of Computer Science, Duke Univ., Durham, NC (US)
Publication Date:
OSTI Identifier:
5001639
Resource Type:
Journal Article
Journal Name:
Journal of Parallel and Distributed Computing; (United States)
Additional Journal Information:
Journal Volume: 11:2; Journal ID: ISSN 0743-7315
Country of Publication:
United States
Language:
English
Subject:
99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; ARRAY PROCESSORS; ALGORITHMS; COMPARATIVE EVALUATIONS; COMPUTER CODES; COST BENEFIT ANALYSIS; MEMORY MANAGEMENT; PARALLEL PROCESSING; ECONOMIC ANALYSIS; ECONOMICS; EVALUATION; MATHEMATICAL LOGIC; PROGRAMMING; 990200* - Mathematics & Computers

Citation Formats

LaRowe, Jr, R P, and Ellis, C S. Page placement policies for NUMA multiprocessors. United States: N. p., 1991. Web. doi:10.1016/0743-7315(91)90117-R.
LaRowe, Jr, R P, & Ellis, C S. Page placement policies for NUMA multiprocessors. United States. https://doi.org/10.1016/0743-7315(91)90117-R
LaRowe, Jr, R P, and Ellis, C S. 1991. "Page placement policies for NUMA multiprocessors". United States. https://doi.org/10.1016/0743-7315(91)90117-R.
@article{osti_5001639,
title = {Page placement policies for NUMA multiprocessors},
author = {LaRowe, Jr, R P and Ellis, C S},
abstractNote = {In many parallel applications, the size of the program's data exceeds even the very large amount of main memory available on large-scale multiprocessors. Virtual memory, in the sense of a transparent management of the main/secondary memory hierarchy, is a natural solution. The replacement, fetch, and placement policies used in uniprocessor paging systems need to be reexamined in light of the differences in the behavior of parallel computations and in the memory architectures of multiprocessors. In this paper the authors investigate the impact of page placement in nonuniform memory access time (NUMA) shared memory MIMD machines. The authors experimentally evaluate several paging algorithms that incorporate different approaches to the placement issue. Under certain workload assumptions, the results show that placement algorithms that are strongly biased toward local frame allocations but are able to borrow remote frames can reduce the number of page faults over strictly local allocation. The increased cost of memory operations due to the extra remote accesses is more than compensated for by the savings resulting from the reduction in demand fetches, effectively reducing the computation completion time for these programs without having adverse effects on the performance of typical NUMA programs. The authors also discuss some early results obtained from an actual kernel implementation of one of the page placement algorithms.},
doi = {10.1016/0743-7315(91)90117-R},
url = {https://www.osti.gov/biblio/5001639}, journal = {Journal of Parallel and Distributed Computing; (United States)},
issn = {0743-7315},
number = ,
volume = 11:2,
place = {United States},
year = {Fri Feb 01 00:00:00 EST 1991},
month = {Fri Feb 01 00:00:00 EST 1991}
}