Effect of CPU Cache Size on OpenMP Computing Performance in Fluid-Film Lubrication Analysis

The growth of the size of cache and the number of processor cores in modern CPUs is the major factor in advancing the computing performance of modern machines. The effect of CPU cache size in multicore computers on performance, however, has attracted little attention in lubrication and engineering analyses. In this study, the effect of cache size on the computational performance of two parallel iterative methods in solving two Reynolds equations is examined. Four computers, with CPU cache size from 4 to 40 MB and the number of processor cores from 4 to 16, were used. The sizes of the numerical grid were selected to simulate large gridwork (256 × 256) to small gridwork (2048 × 2048) tasks. It is found that the size of CPU cache is a major factor influencing the parallel efficiency in using the RBSOR method. On the other hand, the SPSOR method obtains much higher parallel efficiency than the RBSOR for medium-grained tasks, regardless of the size of CPU cache. The use of the SPSOR can, therefore, provide a much better parallel computing performance than the RBSOR in the cases of having a large number of grids or in a system with limited CPU cache.