COST of Graph Processing Using Actors

TL;DR

This paper evaluates the actor model for scalable graph processing, demonstrating that simple actor-based implementations can outperform dedicated systems and serial methods in terms of performance.

Contribution

It provides an empirical analysis showing the effectiveness of actor-based models for scalable graph computations compared to existing approaches.

Findings

Actor-based implementations outperform dedicated graph systems.

Actor models achieve higher performance than serial implementations.

Scalability benefits are realized with simple actor-based designs.

Abstract

Graph processing is an increasingly important domain of computer science, with applications in data and network analysis, among others. Target graphs in these applications are often large, leading to the creation of "big data" systems designed to provide the scalability needed to analyze these graphs using parallel processing. However, researchers have shown that while these systems often provide scalability, they also often introduce overheads that exceed the benefits they provide, sometimes resulting in lower absolute performance than even simple serial implementations. This report studies the viability and performance of the actor model to implement scalable concurrent programs to perform common graph computations. We show that relatively simple actor-based implementations outperform both dedicated graph processing systems and the benchmark serial implementations.