Abstract
As the requirement of large-scale structure analysis becomes stronger, the expectation of parallel computing becomes higher. How to program for parallel processing highly depends on the computer’s architecture, especially memory architecture and parallel interface. This paper discusses parallelisation strategies of general eigenvalue analysis not only based on MPI for distributed memory but also based on OpenMP for shared memory. The performance of MPI, OpenMP, and MPI+OpenMP hybrid parallelised codes are compared on SGI Origin 2000 and Compaq SC series. As a result, we can give that all these approaches work equally well, while the parallelisation with OpenMP has a merit of simplicity; the efficiency of hybrid parallelisation was affected by type of parallel computer and scale of problem.
