High-Performance MPI Library with SR-IOV and SLURM for Virtualized InfiniBand Clusters

In this video from the 2016 OpenFabrics Workshop, DK Panda from Ohio State University presents: High-Performance MPI Library with SR-IOV and SLURM for Virtualized InfiniBand Clusters.

"The MVAPICH2 software libraries have been enabling many HPC clusters during the last 14 years to extract performance, scalability and fault-tolerance using OpenFabrics verbs. As the HPC field is moving to Exascale, many new challenges are emerging to design the next generation MPI, PGAS and Hybrid MPI+PGAS libraries with capabilities to scale to millions of processors while taking advantages of the latest trends in accelerator/co-processor technologies and the features of the OpenFabrics Verbs. In this talk, we will present the approach being taken by the MVAPICH2 project including support for new verbs-level capabilities (DC, UMR, ODP, and offload), PGAS (OpenSHMEM, UPC, CAF and UPC++), Hybrid MPI+PGAS models, tight-integration with NVIDIA GPUs (with GPUDirect RDMA) and Intel MIC, and designs leading to reduced energy consumption. We will also highlight a co-design approach where the capabilities of InfiniBand Network Analysis and Monitoring (INAM) can be used together with the new MPI-T capabilities of the MPI standard to analyze and introspect performance of an MPI program on an InfiniBand cluster and tune it further. We will also present upcoming plans of the MVAPICH2 project to provide support for emerging technologies: OmniPath, KNL and OpenPower."

Learn more: https://www.openfabrics.org/index.php/about-the-2016-ofa-workshop.html

Sign up for our insideHPC Newsletter: http://insidehpc.com/newsletter

Видео High-Performance MPI Library with SR-IOV and SLURM for Virtualized InfiniBand Clusters канала InsideHPC Report