Performance prediction of massively parallel computation by Bayesian inference

TL;DR

This paper introduces a Bayesian inference-based performance prediction method for massively parallel computations, enabling accurate extrapolation of execution time for larger node counts, demonstrated on eigenvalue problems.

Contribution

It presents an improved, generalizable performance prediction approach using Bayesian inference, specifically designed for extrapolating computation times in parallel systems.

Findings

Effective prediction of execution time for larger node counts

Application to real-symmetric generalized eigenvalue problem shows promising results

Method is adaptable to various parallel computations

Abstract

A performance prediction method for massively parallel computation is proposed. The method is based on performance modeling and Bayesian inference to predict elapsed time T as a function of the number of used nodes P (T=T(P)). The focus is on extrapolation for larger values of P from the perspective of application researchers. The proposed method has several improvements over the method developed in a previous paper, and application to real-symmetric generalized eigenvalue problem shows promising prediction results. The method is generalizable and applicable to many other computations.