Hyper-Systolic Processing on the Quadrics: Improving Inter-Processor Communication by Simulated Annealing
暂无分享,去创建一个
Systolic processing is a technique [1] to enlarge the range of aрplications of parallel computers widely, in particular for SIMD and SISAMD (single instruction — single addressing — multiple data) architectures. Its hyper-systolic extension can enhance substantially the inter-processor communication speed. The optimization of the actual hyper-systolic communication layout is a non-trivial task if more than 64 processing nodes are involved. Using Simulated Annealing techniques we are able to improve the recently used regular hyper-systolic bases. We find that the speed of inter-processor communication can be increased by about 20 % on the 25 Gflops Quadrics QH4 (with 512 compute nodes) for n2-problems.
[1] Th. Lippert,et al. HYPER-SYSTOLIC PROCESSING ON APE100/QUADRICS: n2-LOOP COMPUTATIONS , 1996 .
[2] C. D. Gelatt,et al. Optimization by Simulated Annealing , 1983, Science.
[3] Klaus Schilling,et al. Hyper-Systolic Parallel Computing , 1998, IEEE Trans. Parallel Distributed Syst..
[4] Nicolai Petkov,et al. Systolic Parallel Processing , 1992 .
[5] Emile H. L. Aarts,et al. Global optimization and simulated annealing , 1991, Math. Program..