Parallelization of PageRank on multicore processors

Abstract

PageRank is a prominent metric used by search engines for ranking of search results. Page rank of a particular web page is a function of page ranks of all the web pages pointing to this page. The algorithm works on a large number of web pages and is thus computational intensive. The need of hardware is currently served by connecting thousands of computers in cluster. But faster and less complex alternatives to this system can be found in multi-core processors. In this paper, we identify major issues involved in porting PageRank algorithm on Cell BE Processor and CUDA, and their possible solutions. The work is evaluated on three input graphs of different sizes ranging from 0.35 million nodes to 1.3 million. Our results show that PageRank algorithm runs 2.8 times fast on CUDA compared to Xeon dual core 3.0 GHz. © 2012 Springer-Verlag.