A new analysis of Work Stealing with latency

TL;DR

This paper analyzes how communication latency affects the performance of the Work Stealing load balancing algorithm, providing a new predictive expression for expected running time and practical guidelines for processor usage.

Contribution

It introduces a latency-extended model of Work Stealing and derives a new expression for expected makespan, validated through extensive simulation.

Findings

Latency significantly impacts load balancing efficiency.

The new expression predicts performance under various latency conditions.

Guidelines for optimal processor count based on latency are provided.

Abstract

We study in this paper the impact of communication latency on the classical Work Stealing load balancing algorithm. Our paper extends the reference model in which we introduce a latency parameter. By using a theoretical analysis and simulation, we study the overall impact of this latency on the Makespan (maximum completion time). We derive a new expression of the expected running time of a bag of tasks scheduled by Work Stealing. This expression enables us to predict under which conditions a given run will yield acceptable performance. For instance, we can easily calibrate the maximal number of processors to use for a given work/platform combination. All our results are validated through simulation on a wide range of parameters.