Cache-Efficient Approach for Index-Free Personalized PageRank

Personalized PageRank (PPR) measures the importance of vertices with respect to a source vertex. Since real-world graphs are evolving rapidly, PPR computation methods need be index-free and fast. Unfortunately, existing suffer from cache misses. They follow state-of-the-art algorithm that first performs Forward Push (FP) phase subsequently runs random walk Monte-Carlo simulation (MC) phase. Although succeed in reducing misses FP phase, an inefficient data layout limits their performance improvement. Besides, have overlooked MC In this paper, we propose cache-efficient approach accelerates both phases. reorder low overheads. Specifically, utilize Breadth First Search result so near vertex co-located on reordered layout. We perform optimized FP, namely Distance-Extension (DEFP). By preferentially proceeding around vertex, DEFP improves memory access locality. MC, Vertex-Centric Random Walk (VCRW). VCRW aggregates walks at each eliminate redundant for repeatedly obtaining neighbor vertices. prove most can aggregated while maintaining accuracy guarantees. Experimental results show proposed method is up 4.7x faster than outperforms index-oriented under rigorous