DwarvesGraph: A High-Performance Graph Mining System with Pattern Decomposition

This paper presents DwarvesGraph, the first graph mining system that decomposes the target pattern into several subpatterns, and then computes the count of each. The results of the target pattern can be calculated using the subpattern counts with very low additional cost. Despite decomposition-based algorithms have been studied for years, we propose several novel techniques to address key system challenges: 1) a partial-embedding-centric programming model with efficient supports for pattern existence query and advanced graph mining applications such as FSM; 2) an accurate and efficient cost model based on approximate graph mining; 3) an efficient search method to jointly determine the decomposition of all concrete patterns of an application, considering the computation cost and cross-pattern computation reuse; and 4) the partial symmetry breaking technique to eliminate redundant enumeration for each subpattern while preserving equivalence of computation. Our experiments show that DwarvesGraph is significantly faster than all existing state-of-the-art systems and provides a novel and viable path to scale to large patterns.