Improving Performance Models for Irregular Point-to-Point Communication

TL;DR

This paper introduces an enhanced performance model for irregular point-to-point communication that accounts for queue search time and network contention, improving accuracy in predicting communication costs in parallel applications.

Contribution

It presents a novel node-aware performance model that extends traditional models to better capture irregular communication costs in parallel computing.

Findings

Model shows improved accuracy over traditional models.

Effectively predicts communication costs in irregular matrix operations.

Incorporates network contention and queue search time into the model.

Abstract

Parallel applications are often unable to take full advantage of emerging parallel architectures due to scaling limitations, which arise due to inter-process communication. Performance models are used to analyze the sources of communication costs. However, traditional models for point-to-point communication fail to capture the full cost of many irregular operations, such as sparse matrix methods. In this paper, a node-aware based model is presented. Furthermore, the model is extended to include communication queue search time as well as an additional parameter estimating network contention. The resulting model is applied to a variety of irregular communication patterns throughout matrix operations, displaying improved accuracy over traditional models.