Enhancing Interval-centric Distributed Computing to Support Incremental Graphs

Real-world graphs like social networks, fintech transactions and the World Wide Web (WWW) are dynamic in nature, and have addition and deletion of vertices and edges. The Interval-centric model (ICM) was proposed to intuitively design and execute distributed algorithms over temporal graphs having a lifespan on vertices, edges and their attributes. The temporal graphs are available a priori. The Windowed ICM (WICM) model further optimized its performance by avoiding the loading of the entire temporal graph in-memory. Here, we extend the WICM model to incrementally operate on dynamic graphs whose vertex and edge updates arrive over time. We achieve this by decoupling the receipt and application of updates to the temporal graph, and the computation on the graph using 3 strategies – JITM, DITM and AITM. We compare our results with the native WICM and strawman baseline. Our evaluation on the Reddit graph shows $\approx 40$% better performance without compromising on correctness