Introducing ScaleGraph: An X10 library for billion scale graph analytics

Abstract

Highly Productive Computing Systems (HPCS) and PGAS languages are considered as important ways in achieving the exascale computational capabilities. Most of the current large graph processing applications are custom developed using non-HPCS/PGAS techniques such as MPI, MapReduce. This paper introduces Scale-Graph, an X10 library targeting billion scale graph analysis scenarios. Compared to non-PGAS alternatives, ScaleGraph defines concrete, simple abstractions for representing massive graphs. We have designed ScaleGraph from ground up considering graph structural property analysis, graph clustering and community detection. We describe the design of the library and provide some initial performance evaluation results of the library using a twitter graph with 1.47 billion edges. Copyright © 2012 ACM.