Topology-Aware MPI and Job Scheduling

Purpose

HPC scheduling systems are generally not adjusted for the underlying topology of the computers on which they run. As a result, these systems compute less efficiently than an intelligent decision-making system would. Supported by an STCI grant from the NSF, researchers at TACC are trying to endow the Oracle Grid Engine (OGE), which runs on Ranger, with dynamic topology information to improve its performance.

Funding Source(s)

  • Topology Aware MPI
Bill Barth

Director of High Performance Computing
bbarth@tacc.utexas.edu | 512-475-9411