Title :
Building cost effective high performance computing environment via PCI express
Author_Institution :
CREST, Chuo Univ.
Abstract :
Building a scalable and low cost parallel computing environment is becoming a critical matter for scientific computing that requires the repetitive solutions of large linear systems. In this study, a PCI Express based PC cluster is built for the scalable implementation of parallel iterative linear solvers, and the performance and bottlenecks are evaluated. The results show that the communication bandwidth of the internode network is the most important factor for the scalable implementation of parallel sparse linear algebra operations
Keywords :
bandwidth allocation; parallel processing; performance evaluation; peripheral interfaces; workstation clusters; InfiniBand; NUMA systems; Opteron processor; PC cluster; PCI express; communication bandwidth; high performance computing environment; internode network; linear systems; parallel computing environment; parallel iterative linear solvers; parallel sparse linear algebra operations; scientific computing; Bandwidth; Concurrent computing; Costs; Gradient methods; High performance computing; Large-scale systems; Libraries; Linear algebra; Parallel processing; Sparse matrices; Express; InfiniBand; NUMA systems; Opteron processor; PCI; sparse linear algebra operations;
Conference_Titel :
Parallel Processing Workshops, 2006. ICPP 2006 Workshops. 2006 International Conference on
Conference_Location :
Columbus, OH
Print_ISBN :
0-7695-2637-3
DOI :
10.1109/ICPPW.2006.27
