The Community for Technology Leaders
Parallel and Distributed Processing Symposium, International (2007)
Long Beach, CA, USA
Mar. 26, 2007 to Mar. 30, 2007
ISBN: 1-4244-0909-8
pp: 391
James Dinan , Dept. of Comp. Sci. and Engineering, The Ohio State University, Columbus, OH 43221. dinan@cse.ohio-state.edu
Stephen Olivier , Dept. of Computer Science, Univ. of North Carolina at Chapel Hill, Chapel Hill, NC 27599. olivier@cs.unc.edu
Gerald Sabin , Dept. of Comp. Sci. and Engineering, The Ohio State University, Columbus, OH 43221. sabin@cse.ohio-state.edu
Jan Prins , Dept. of Computer Science, Univ. of North Carolina at Chapel Hill, Chapel Hill, NC 27599. prins@cs.unc.edu
P. Sadayappan , Dept. of Comp. Sci. and Engineering, The Ohio State University, Columbus, OH 43221. saday@cse.ohio-state.edu
Chau-Wen Tseng , Dept. of Computer Science, Univ. of Maryland at College Park, College Park, MD 20742. tseng@cs.umd.edu
ABSTRACT
This paper examines MPI's ability to support continuous, dynamic load balancing for unbalanced parallel applications. We use an unbalanced tree search benchmark (UTS) to compare two approaches, 1) work sharing using a centralized work queue, and 2) work stealing using explicit polling to handle steal requests. Experiments indicate that in addition to a parameter defining the granularity of load balancing, message-passing paradigms require additional parameters such as polling intervals to manage run-time overhead. Using these additional parameters, we observed an improvement of up to 2X in parallel performance. Overall we found that while work sharing may achieve better peak performance on certain workloads, work stealing achieves comparable if not better performance across a wider range of chunk sizes and workloads.
INDEX TERMS
null
CITATION

J. Prins, S. Olivier, J. Dinan, P. Sadayappan, C. Tseng and G. Sabin, "Dynamic Load Balancing of Unbalanced Computations Using Message Passing," 2007 IEEE International Parallel and Distributed Processing Symposium(IPDPS), Rome, 2007, pp. 391.
doi:10.1109/IPDPS.2007.370581
89 ms
(Ver 3.3 (11022016))