DMT: A Flexible and Versatile Selectivity Estimation Approach for Graph Query

Efficient and accurate selectivity estimation in graph-structured data, specifically for complex branching path query, is becoming a challenging and all-important problem for query performance optimization. Precise and flexible statistics summarization about graph-structured data plays a crucial role for graph query selectivity estimation. We propose DMT, Dynamic Markov Table, which is a dynamic graph summarization based on Markov Table by applying flexible combination of 4 Optimized Rules which investigate local forward and backward inclusions. The efficient DMT construction algorithm DMTBuilder and DMT-based statistical methods are introduced for selectivity estimations of various graph queries. Our extensive experiments demonstrate the advantages in accuracy and scalability of DMT by comparing with previously known alternative, as well as the preferred Optimized Rules that would favor different situations.