Isolate-Set-Based In-Memory Parallel Subgraph Matching Framework

Subgraph matching is a classical graph task, which finds all subgraphs that are isomorphic to the query graph in a labeled data graph. According to the theoretical basis, there are three types of subgraph matching methods: exploration-based methods, state-space representation methods, and multi-way join methods. Each of these methods has its advantages, but their common feature is a large amount of computation. In order to improve the computation efficiency, some representative subgraph matching methods have their own parallel optimization methods, but these parallel optimization methods are not universal. To overcome the drawbacks and improve the efficiency of parallel computing, we propose an isolate-set-based parallel subgraph matching framework. The parallel framework can flexibly adapt different subgraph matching methods without changing the sequential methods themselves. We partition the search tree into several decoupled subtrees and prune the subtrees to reduce computation overheads. And then, parallel subgraph matching is implemented on the decoupled pruned subtrees. To demonstrate the proposed method’s effectiveness empirically, we parallelize several representative subgraph matching methods in our framework and compare their performance with the sequential methods. The results show that our parallel framework is adapted to representative sequential methods, outperforms the sequential methods by a large margin, and achieves a speedup of 10x-50x on a 64-core x86 machine.