An efficient partition-based parallel PageRank algorithm

PageRank becomes the most well-known re-ranking technique of the search results. By its iterative computational nature, the computation takes much computing time and resource. Researchers have then devoted much attention in studying an efficient way to compute the PageRank scores of a very large Web graph. However, only a few of them focus on large-scale PageRank computation using parallel processing techniques. In this paper, we propose a partition-based parallel PageRank algorithm that can efficiently run on a low-cost parallel environment like the PC cluster. For comparison, we also study the other two known techniques, as well as propose an analytical discussion concerning I/O and synchronization cost, and memory usage. Experimental results with two Web graphs synthesized from the .TH domain and the Stanford WebBase project are very promising.