Parallel SCC Detection Based on Reusing Warps and Coloring Partitions on GPUs

Abstract

Detecting strongly connected components (SCCs) in directed graphs is crucial for analyzing the structure of graphs. It can accelerate the process of large-scale graphs with applying GPUs to graph computation. However, GPU-based SCC detection still faces the challenge from graphs, especially real-world graphs, skewing and strict synchronization requirement. In this paper, we proposed a new paradigm of identifying SCCs with reusing warps and coloring partitions. The scheme of reusing warps assigns multiple vertices to each virtual warp at one time, which greatly reduces the number of warps that are applied but assigned few tasks. Furthermore, we proved that the partitions formed by coloring independently contains all SCCs in them, so we use coloring to identify more partitions and parallel detect SCCs in each partition to get more SCCs. Finally, the colored partitions are deeply combined with the following traversal to optimize the algorithm. We conduct extensive theoretical and experimental analysis to demonstrate the efficiency and accuracy of our approach. The experimental results expose that our approach can achieve 7.23\(\times \), 30.55\(\times \), 1.75\(\times \) and 1.26\(\times \) speedup for SCC detection using NVIDIA K80 compared with algorithms of Tarjan, Barnat, Hong and Slota respectively.