ARMing up with MVAPICH, a performance comparison of different MPI implementations on an ARM HPC system

Activity: Academic talk or presentation typesInvited talk

Description

With a variety of new hardware technologies becoming available for HPC it is an exciting time at the moment for our community. One, potentially very important, future player in the HPC space will be ARM and there are already a number of ARM CPU based HPC system already available. In this talk I will describe work we have done in exploring the performance properties of MVAPICH, OpenMPI and MPT on one of these systems, Fulhame, which is an HPE Apollo 70-based system with 64 nodes of Cavium ThunderX2 ARM processors and Mellanox InfiniBand interconnect. In order to take advantage of these systems most effectively, it is very important to understand the performance that different MPI implementations can provide and any further opportunities to optimise these. Therefore, starting with the OSU benchmarks I will explore the different performance properties of these technologies on Fulhame and other systems for comparison, before then moving onto a number of more substantial applications.
Period21 Aug 2019
Event title7th Annual MVAPICH User Group (MUG) Meeting
Event typeConference
LocationOhio, United StatesShow on map
Degree of RecognitionInternational