Distributing an Exact Algorithm for Maximum Clique: maximising the costup

TL;DR

This paper presents a simple modification to a state-of-the-art maximum clique algorithm, enabling its distribution over multiple machines to achieve significant speedups with minimal development effort.

Contribution

The authors introduce a straightforward method to distribute an existing exact maximum clique algorithm across multiple machines, achieving high parallel speedups.

Findings

Speedups of an order of magnitude with 25+ machines

Minimal development effort required for distribution

Effective distribution on large hard benchmarks

Abstract

We take an existing implementation of an algorithm for the maximum clique problem and modify it so that we can distribute it over an ad-hoc cluster of machines. Our goal was to achieve a significant speedup in performance with minimal development effort, i.e. a maximum costup. We present a simple modification to a state-of-the-art exact algorithm for maximum clique that allows us to distribute it across many machines. An empirical study over large hard benchmarks shows that speedups of an order of magnitude are routine for 25 or more machines.