An efficient algorithm for density-balanced partitioning in distributed pagerank

Google's PageRank is the most notable approach for web search ranking. In general, web pages are represented by web-link graph; a web-page is represented by a node, and a link between two pages is represented by an edge. In particular, it is not efficient to perform PageRank of a large web-link graph in a single computer. Distributed systems, such as P2P, are viable choices to address such limitation. In P2P-based PageRank, each computational peer contains a partial web-link graph, i.e., a sub-graph of the global web-link graph, and its PageRank is computed locally. The convergence time of a PageRank calculation is affected by the web-link graph density, i.e., the ratio of the number of edges to the number of nodes, such that if a web-link graph has high density, it will take longer time to converge. As the execution time to compute the P2P-based web ranking is influenced by the execution time of the slowest peer to compute the local ranking, the density-balanced local web-link graph partitioning can be highly desirable. This paper addresses a density-balanced partitioning problem and proposes an efficient algorithm for the problem. The experiment results show that the proposed algorithm can effectively partition graph into density-balanced sub with an acceptable cost.