Treffer: Benefits of MPI Sessions for GPU MPI applications

International audience ; Heterogeneous supercomputers are now considered the most valuable solution to reach the Exascale. Nowadays, we can frequentlyobserve that compute nodes are composed of more than one GPUaccelerator. Programming such architectures efficiently is challenging.MPI is the defacto standard for distributed computing. CUDAaware libraries were introduced to ease GPU inter-nodes communications. However, they induce some overhead that can degradeoverall performances. MPI 4.0 Specification draft introduces theMPI Sessions model which offers the ability to initialize specificresources for a specific component of the application.In this paper, we present a way to reduce the overhead inducedby CUDA-aware libraries with a solution inspired by MPI Sessions.In this way, we minimize the overhead induced by GPUs in an MPIcontext and allow to improve CPU + GPU programs efficiency. Weevaluate our approach on various micro-benchmarks and someproxy applications like Lulesh, MiniFE, Quicksilver, and Cloverleaf.We demonstrate how this approach can provide up to a 7x speedupcompared to the standard MPI model.