Evaluation of the Partitioned Global Address Space (PGAS) model for an inviscid Euler solver

摘要

In this paper we evaluate the performance of Unified Parallel C (whichimplements the partitioned global address space programming model) using anumerical method that is widely used in fluid dynamics. In order to evaluatethe incremental approach to parallelization (which is possible with UPC) andits performance characteristics, we implement different levels of optimizationof the UPC code and compare it with an MPI parallelization on four differentclusters of the Austrian HPC infrastructure (LEO3, LEO3E, VSC2, VSC3) and on anIntel Xeon Phi. We find that UPC is significantly easier to develop in comparedto MPI and that the performance achieved is comparable to MPI in mostsituations. The obtained results show worse performance (on VSC2), competitiveperformance (on LEO3, LEO3E and VSC3), and superior performance (on the IntelXeon Phi).