DETER: Streaming Graph Partitioning via Combined Degree and Cluster Information

Cong Hu, Jiang Zhong, Qi Li, Qing Li

Abstract: Efficient graph partitioning plays an important role in distributed graph processing systems with the rapid growth of the scale of graph data. The quality of partitioning affects the performance of systems greatly. However, most existing vertex-cut graph partitioning algorithms only focused on degree information and ignored the cluster information of a coming edge when assigning edges. It is beneficial to assign an edge to a partition with more neighbors because keeping a dense subgraph in one partition would reduce the communication cost. In this paper, we propose DETER, an efficient vertex-cut streaming graph partitioning algorithm that takes both degree and cluster information into account when assigning an edge to one partition. Our evaluations suggest that DETER algorithm owns the ability to efficiently partition large graphs and reduce communication cost significantly compared to state-of-the-art graph partitioning algorithms.