Optimizing CPU cache performance for Pregel-like graph computation

In-memory graph computation systems have been used to support many important applications, such as PageRank on the web graph and social network analysis. In this paper, we study the CPU cache performance of graph computation. We have implemented a graph computation system, called GraphLite, in C/C++ based on the description of Pregel. We analyze the CPU cache behavior of the internal data structures and operations of graph computation. Then we exploit CPU cache prefetching techniques to improve the cache performance. Real machine experimental results show that our solution achieves 1.9-2.2x speedups compared to the baseline implementation.