Issue No. 10 - October (1994 vol. 5)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/71.313122
<p>We present new methods for load balancing of unstructured tree computations onlarge-scale SIMD machines, and analyze the scalability of these and other existingschemes. An efficient formulation of tree search on an SIMD machine consists of twomajor components: a triggering mechanism, which determines when the search spaceredistribution must occur to balance the search space over processors, and a scheme toredistribute the search space. We have devised a new redistribution mechanism and anew triggering mechanism. Either of these can be used in conjunction with triggering andredistribution mechanisms developed by other researchers. We analyze the scalability ofthese mechanisms and verify the results experimentally. The analysis and experimentsshow that our new load-balancing methods are highly scalable on SIMD architectures.Their scalability is shown to he no worse than that of the best load-balancing schemeson MIMD architectures. We verify our theoretical results by implementing the 15-puzzleproblem on a CM-2 SIMD parallel computer.</p>
Index Termsresource allocation; parallel machines; trees (mathematics); search problems; parallelarchitectures; performance evaluation; unstructured tree search; SIMD parallelcomputers; load balancing; unstructured tree computations; large-scale SIMD machines;scalability; tree search; SIMD machine; triggering mechanism; search spaceredistribution; load-balancing methods; load-balancing schemes; MIMD architectures;15-puzzle problem; CM-2 SIMD parallel computer
G. Karypis and V. Kumar, "Unstructured Tree Search on SIMD Parallel Computers," in IEEE Transactions on Parallel & Distributed Systems, vol. 5, no. , pp. 1057-1072, 1994.