SGP: A Parallel Computing Framework for Supporting Distributed Structural Graph Clustering

Abstract

Structural graph clustering is an important problem in the domain of graph data management. Given a large graph G, structural graph clustering is to assign vertices to clusters where vertices in the same cluster are densely connected to each other and vertices in different clusters are loosely connected to each other. Due to its importance, many algorithms have been proposed to study this problem. However, no effort focuses on the distributed graph environment. In this paper, we propose a parallel computing framework named SGP (short for Statistics-based Graph Partition) to support large graph clustering under distributed environment. We first use historical clustering information to partition graph into a group of clusters. Based on the partition result, we can properly assign vertexes to different nodes based on connection relationship among vertex. When a clustering request is submitted, we can use properties leading by the partition for efficiently clustering. Finally, we conduct extensive performance studies on large real and synthetic graphs, which demonstrate that our new approach could efficiently support large graph clustering under distributed environment.