Cyclosa - Redundancy-Free Graph Pattern Mining via Set Dataflow
Explore a conference talk from USENIX ATC '23 that introduces Cyclosa, a novel dataflow-based graph pattern mining framework. Learn how this innovative approach eliminates redundancies in pattern-centric mining systems by leveraging computation similarity. Discover the three key features of Cyclosa: a set dataflow representation that optimizes scheduling and identifies potential redundancies, a parallel execution engine that decouples data access and computations for efficient result sharing, and a memory-friendly data management substrate that automatically manages high-reuse computation results. Understand how Cyclosa outperforms state-of-the-art systems like GraphPi and SumPA, achieving up to 16.28× and 5.52× performance improvements respectively. Gain insights into the importance of graph pattern mining across various fields and how Cyclosa addresses the challenges of recurrent set intersections in existing systems.