Partition-Oriented Subgraph Matching on GPU

Abstract

Subgraph isomorphismis a well known NP-hard problem that finds all the matched subgraphs of a query graph in a large data graph. The state-of-the-art GPU-based solution is the vertex-oriented joining strategy, which is proposed by GSI. It effectively solves the problem of parallel write conflicts by taking vertices as processing units. However, this strategy might result in load-imbalance and redundant memory transactions when dealing with dense query graph. In this paper, we design a new storage structure Level-CSR and a new partition-oriented joining strategy. To avoid the influence of vertices with large degrees, we divide the dense vertices in traditional CSR into several GPU-friendly tasks and store them in Level-CSR. Then, an efficient execution strategy is designed based on the partitioned tasks. The partition strategy can improve the load imbalance caused by the irregularity of real-world graphs, and further reduce the redundant global memory access caused by the redundant neighbor set accessing. Besides, to further improve the performance, we propose a well-directed filtering strategy by exploiting a property of real-world graphs. The experiments show that compared with the state-of-the-art GPU based solutions, our approach can effectively reduce the number of unrelated candidates, minimize memory transactions, and achieve load balance between processors.