Modelagem assíncrona do Page Rank

Abstract

In this work, we present an exact distributed solution for the problem of ranking the most popular vertices of a graph. In recent years, the amount of WWW links has become increasing, so sequential or just parallel algorithms are no longer feasible as they run on just one machine. In this work, we redesign the PageRank algorithm, proposed by Google’s founders, so that it runs efficiently under machine clusters. Bringing an innovative proposal in relation to the literature with asynchronous modeling. The solution uses the Java Cá & Lá (JCL) middleware and it has been tested against one of the industry leaders-the Apache GraphX. The results using synthetic and real instances demonstrated that our solution, named JCL PageRank, was 10x faster than GraphX. We have performed experiments with small and big machines in two different cluster configurations and the runtime results were similar. Both JCL and GraphX solutions calculate the PageRank values for all vertices of a graph, thus without approximations. Personalized PageRank calculus is also performed by both GraphX and JCL PageRank solutions.