Hybrid MPI/Open-MP acceleration approach for high-order schemes for CFD

Abstract

The paper presents hybrid MPI+OpenMP (message Passing Interface/Open Multi-Processor) used for paralleled programs including high-order compact method. The main tools used to implement parallelism in computations are OpenMP and MPI which differ in terms of memory they are based on. OpenMP works on shared-memory and MPI on distributed-memory whereas hybrid model is based on combination of those methods. Tests performed and described in this paper present significant advantages provided by hybrid MPI/OpenMP. Testing computations needed for verifying possibilities of MPI, Open-MP and Hybrid of both are carried out using an academic high-order SAILOR solver. Obtained results seem to be very promising to accelerate simulations for fluid flows as well as for application using high order methods. The proposed approach was tested up to 96 cores with up to 4 nodes.