Multithreading Heterogeneous Graph Aggregation

Kai Zou,Xike Xie,Haoyun Li,X. Sean Wang

DOI: https://doi.org/10.1109/tkde.2023.3320127

IF: 9.235

2023-01-01

IEEE Transactions on Knowledge and Data Engineering

Abstract:Towards building online analytical services on big heterogeneous graphs, we study the problem of the multithreading graph aggregation. The purpose is to exploit the thread-level parallelism for accelerating the graph aggregation process, which is both data and computation intensive. We identify the sources of parallelization latency caused by multifarious factors, including data distributions and contentions, uneven workload assignments, logical aggregation plan obstructions, etc. To cope with these problems, we investigate a parallelization solution for graph aggregation with a number of threads packaged as threadblocks, categorize the parallelization latency as the thread-level and threadblock-level latency, and propose a series of optimization techniques for alleviating or eliminating the latency on different levels. The solution supports different aggregate functions, scales up to large number of threads, and scales out to big heterogeneous graphs. Experiments on real datasets show that our solution achieves up to 60x acceleration with 256 threads compared to the non-parallelized solution.