Speed-up and efficiency of parallelized Monte Carlo integration on homogeneous and heterogeneous clusters

Authors

Ryan Carlos M Tabernilla; John Kevin R Sanchez; Francis N. C. Paraan

Source

Proceedings of the 32nd Samahang Pisika ng Pilipinas Physics Congress, University of the Philippines Diliman, Quezon City, SPP2014-2C-6 (2014).

Abstract

We evaluate the performance of parallelized Monte Carlo integration algorithms on homogeneous and heterogeneous clusters of the Structure and Dynamics Laboratory. In this study, intrinsic pseudorandom number generators for Fortran and Python were used and parallelization achieved by MPI libraries. On an example using 10^9 samples, the parallelized Python code proved to be scalable on a cluster of up to 44 processors. The Fortran parallelized code performed less well on scalability but had a much shorter execution time. It was also observed that the overhead cost of parallelization saturates as the number of processors used increased.