STIC-D: algorithmic techniques for efficient parallel pagerank computation on real-world graphs

Abstract

Computing metrics on nodes of a graph is an essential step in understanding the properties of the graph. Pagerank is one such metric that is popular and is being used to measure the importance of nodes in not only web graphs but also in social networks, biological networks, road networks, and the like. The core of the computation of pagerank can be seen as an iterative approach that updates the pageranks of nodes until the values converge. However, as real-world graphs such as road networks and the web have a large size, one needs to design efficient techniques to address the challenges of scale. In addition to parallelism that can be exploited, it is important to also look for specific properties of graphs and their impact on the algorithm. In this paper, we present four algorithmic techniques that optimize the pagerank computation on real-world graphs. The techniques are presented with the aim of exploiting the nature of the real-world graphs and eliminating redundancies in the pagerank computation. Our techniques also have the advantage that with little extra effort one can quickly identify which of the techniques will be suitable for a given input graph. We implement our algorithm on an Intel i7 980x CPU running 12 threads using OpenMP Version 3.0. We study our techniques on four classes of real-world graphs: web graphs, social networks, citation and collaboration networks, and road networks. Our implementation achieves an average speedup of 32% compared to a baseline implementation.