BENU: Distributed Subgraph Enumeration with Backtracking-Based Framework

Abstract

Given a small pattern graph and a large data graph, the task of subgraph enumeration is to find all subgraphs of the data graph that are isomorphic to the pattern graph. When the data graph is dynamic, the task of continuous subgraph enumeration is to detect the changes in the matching results caused by the edge updates at each time step. The two tasks are fundamental in many graph analysis applications. The state-of-the-art distributed methods solve them via distributed multi-way join. However, they are inefficient in communication since they have to shuffle partial matching results during the join. The partial matching results may be much larger than the data graph itself. To overcome the drawback, we develop the BENU framework for distributed subgraph enumeration. Given a data graph, BENU generates a group of local search tasks that follow a backtracking-based execution plan to enumerate subgraphs of the pattern graph. BENU executes the tasks in parallel. The tasks query the data graph stored in a distributed database on demand, avoid shuffling partial matching results. To handle dynamic data graphs, we propose the concept of incremental pattern graphs. We solve the continuous subgraph enumeration via enumerating incremental pattern graphs in the data graph snapshots at each time step. We extend BENU into S-BENU to enumerate them efficiently. We develop implementations for BENU and S-BENU with two optimization techniques. The extensive experiments show that BENU and S-BENU are scalable. They outperform the state-of-the-art by up to one and two orders of magnitude, respectively.Index Terms-backtracking-based framework, continuous subgraph matching, distributed graph querying, subgraph isomorphism, subgraph matching. !

A Survey on Distributed Graph Pattern Matching in Massive Graphs

“…We discuss this category more in detail in Section 4 after introducing the different concepts and programming models for distributed graph processing in Section 3. However, as an exception, we present here the four works QFrag [71], CECI [3], PSM [16] and BENU [92]. First, QFrag and CECI replicate the data graph on multiple machines where each processor runs ISO search on different parts of the data graph.…”

“…Finally, BENU [92] also uses backtracking and decomposes the search tree into regions to be processed by different workers. The workers use adjacency lists of the data graph to enumerate possible embeddings through the application of set intersections.…”

BFS based distributed algorithm for parallel local directed sub-graph enumeration

“…• Enumeration approaches (Park et al [2016], Wang et al [2019]) are counts by enumerating all EFFICIENT GPU BASED LOCAL MOTIF COUNTING 3 of 16 matches of p in G. This is obviously the most computationally expensive method, but it is not limited to specific types of graphs in undirected graphs.…”

HUGE: An Efficient and Scalable Subgraph Enumeration System

“…With the growth in graph size nowadays [67] and the NP-hardness [27] of subgraph enumeration, researchers have devoted enormous efforts into developing efficient and scalable algorithms in the distributed context [5,45,46,60,63,66,71,80,84]. Motivations.…”

“…The efficiency and scalability of distributed subgraph enumeration are jointly determined by three perspectives: computation, communication and memory management [47,66]. However, existing works [5,46,66,84] 1 fail to demonstrate satisfactory performance for all three perspectives. To verify, we conduct an initial experiment by running the square query (□) over the popular benchmark graph LJ [47].…”