FrogWild! - Fast PageRank Approximations on Graph Engines

We propose FrogWild, a novel algorithm for fast approxi-mation of high PageRank vertices, geared towards reducingnetwork costs of running traditional PageRank algorithms.Our algorithm can be seen as a quantized version of poweriteration that performs multiple parallel random walks overa directed graph. One important innovation is that we in-troduce a modification to the GraphLab framework thatonly partially synchronizes mirror vertices. This partialsynchronization vastly reduces the network traffic generatedby traditional PageRank algorithms, thus greatly reducingthe per-iteration cost of PageRank. On the other hand,this partial synchronization also creates dependencies be-tween the random walks used to estimate PageRank. Ourmain theoretical innovation is the analysis of the correla-tions introduced by this partial synchronization process anda bound establishing that our approximation is close to thetrue PageRank vector.We implement our algorithm in GraphLab and compareit against the default PageRank implementation. We showthat our algorithm is very fast, performing each iteration inless than one second on the Twitter graph and can be up to7× faster compared to the standard GraphLab PageRankimplementation.