Comparing commodity SMP system software with a matrix multiplication benchmark

摘要

Commodity symmetric multiprocessors (SMPs), though originally intended for transaction processing, because of their availability, are now used for numerical analysis applications in SMP clusters. Dense matrix multiplication is a suitable benchmark as it exposes memory access issues. Various block-based algorithms are compared across Windows 2000 and Linux variants with gcc compiler versions. Differences in compilers may explain differential performance between algorithms.