Optimizing the Performance of SPEC MPI with Skylake and Omni-Path
Optimizing the Performance of SPEC MPI with Skylake and Omn-Path
Tuning the Eager threshold has a significant impact on application performance by avoiding the synchronization of rendezvous protocol and thus yielding better communication computation overlap.
Library Version: MVAPICH2 2.3rc1
Runtime Flags:
| milc | MV2_SMP_EAGERSIZE=256000 |
| leslie3d | None |
| pop2 | None |
| lammps | MV2_SMP_EAGERSIZE=1024000 |
| wrf2 | MV2_SMP_EAGERSIZE=128000 |
| GAPgeofem | None |
| tera_tf | MV2_SMP_EAGERSIZE=128000 |
| lu | MV2_SMP_EAGERSIZE=1024000 |
System Details:
| CPU Model | Intel Xeon Platinum 8160 |
| CPU Core Info | 2 x 24 @ 2.1 GHz |
| IB Card | 100Gb/sec Intel Omni-Path (OPA) |
| IB Switch | Full Fat Tree 100Gb/sec Intel Omni-Path with 6 core switches |
Tags: MILC SpecMPI Parallel Ocean Program 2 (pop2) LAMMPS Molecular Dynamics Simulator WRF2 GAPgeofem TERA_TF LU
Submitted by Amit Ruhela @ OSU
Last Modified March 6, 2018, 1:53 p.m.
