English

Understanding performance variability in standard and pipelined parallel Krylov solvers

Mathematical Software 2021-03-24 v1

Abstract

In this work, we collect data from runs of Krylov subspace methods and pipelined Krylov algorithms in an effort to understand and model the impact of machine noise and other sources of variability on performance. We find large variability of Krylov iterations between compute nodes for standard methods that is reduced in pipelined algorithms, directly supporting conjecture, as well as large variation between statistical distributions of runtimes across iterations. Based on these results, we improve upon a previously introduced nondeterministic performance model by allowing iterations to fluctuate over time. We present our data from runs of various Krylov algorithms across multiple platforms as well as our updated non-stationary model that provides good agreement with observations. We also suggest how it can be used as a predictive tool.

Keywords

Cite

@article{arxiv.2103.12067,
  title  = {Understanding performance variability in standard and pipelined parallel Krylov solvers},
  author = {Hannah Morgan and Patrick Sanan and Matthew G. Knepley and Richard Tran Mills},
  journal= {arXiv preprint arXiv:2103.12067},
  year   = {2021}
}

Comments

18 pages, 12 figures

R2 v1 2026-06-24T00:26:24.286Z