Dynamic Graph Repartitioning: From Single Vertex to Vertex Group

With the increase of large graph data arising in applications like Web, social network, knowledge graph, and so on, there is a growing need for partitioning and repartitioning large graph data in graph data systems. However, the existing graph repartitioning methods are known for poor efficiency in the dynamic environment. In this paper, we devise an efficient lightweight method to identify and move the candidate vertices to achieve graph repartitioning in the dynamic environment. Different from previous approaches that just focus on the case of moving a single vertex as a basic unit, we show that the movement of some closely connected vertices as a group can further improve the quality of graph repartitioning result. We conduct experiments on a large set of real and synthetic graph data sets, and the results showed that the proposed method is more efficient comparing with existing method in several aspects.