Fast Tuning of Intra-Cluster Collective Communications

TL;DR

This paper explores optimizing intra-cluster collective communications to enhance overall system efficiency, proposing a performance prediction-based approach to select the best communication strategy for different network environments.

Contribution

It evaluates a simple optimization method that compares implementation strategies using communication models to select the most suitable strategy per network environment.

Findings

Performance prediction effectively guides strategy selection.

Different implementation strategies perform variably across environments.

Optimized intra-cluster communication improves overall efficiency.

Abstract

Recent works try to optimise collective communication in grid systems focusing mostly on the optimisation of communications among different clusters. We believe that intra-cluster collective communications should also be optimised, as a way to improve the overall efficiency and to allow the construction of multi-level collective operations. Indeed, inside homogeneous clusters, a simple optimisation approach rely on the comparison from different implementation strategies, through their communication models. In this paper we evaluate this approach, comparing different implementation strategies with their predicted performances. As a result, we are able to choose the communication strategy that better adapts to each network environment.