OLPGP: An Optimized Label Propagation-Based Distributed Graph Partitioning Algorithm

Abstract

One of the concepts that have attracted attention since entering the big data era is graph-structured data. Distributed systems for graph analysis are widely used to process large graphs. Graph partitioning is critical in parallel and distributed graph processing systems because it can balance the computational load and reduce communication load. An efficient graph partitioning algorithm can significantly improve the performance of large-scale graph data analysis and processing. In this paper, we propose a new Optimized Label Propagation-based distributed Graph Partitioning algorithm (OLPGP). OLPGP optimizes the label propagation algorithm and considers the differences between nodes. To improve computational efficiency, we implement OLPGP on the open-source distributed graph processing framework Spark GraphX. Conducted experiments on real-world networks indicate that OLPGP is scalable and achieves higher partition quality than the state-of-the-art label propagation-based graph partitioning algorithms.