SOWalker - An I-O-Optimized Out-of-Core Graph Processing System for Second-Order Random Walks

Explore an innovative I/O-optimized out-of-core graph processing system for second-order random walks in this 18-minute conference talk from USENIX ATC '23. Discover how SOWalker, developed by researchers from Wuhan National Laboratory for Optoelectronics at Huazhong University of Science and Technology, addresses the scalability challenges of random walks in real-world scenarios. Learn about the system's key features, including a walk matrix to reduce non-updatable walks and eliminate useless walk I/Os, a benefit-aware I/O model for improved I/O utilization, and a block set-oriented walk updating scheme to boost walk updating rates. Gain insights into how SOWalker outperforms state-of-the-art random walk systems, achieving up to 10.2× performance speedups compared to GraphWalker and GraSorw.