Making Applications Faster by Asynchronous Execution: Slowing Down Processes or Relaxing MPI Collectives

TL;DR

This paper explores how intentionally slowing down MPI processes or relaxing collective operations can improve performance in memory-bound parallel programs on HPC clusters, challenging conventional views on noise and synchronization.

Contribution

It introduces the concept of leveraging controlled delays and asynchronous MPI communication to enhance performance, supported by visualization tools and experiments on various benchmarks.

Findings

Deliberate delays can improve performance under certain conditions

Noise can be beneficial rather than detrimental in parallel execution

Phase-space graphs effectively visualize complex program dynamics

Abstract

Comprehending the performance bottlenecks at the core of the intricate hardware-software interactions exhibited by highly parallel programs on HPC clusters is crucial. This paper sheds light on the issue of automatically asynchronous MPI communication in memory-bound parallel programs on multicore clusters and how it can be facilitated. For instance, slowing down MPI processes by deliberate injection of delays can improve performance if certain conditions are met. This leads to the counter-intuitive conclusion that noise, independent of its source, is not always detrimental but can be leveraged for performance improvements. We employ phase-space graphs as a new tool to visualize parallel program dynamics. They are useful in spotting certain patterns in parallel execution that will easily go unnoticed with traditional tracing tools. We investigate five different microbenchmarks and applications on different supercomputer platforms: an MPI-augmented STREAM Triad, two implementations of Lattice-Boltzmann fluid solvers, and the LULESH and HPCG proxy applications.