Fast Distributed PageRank Computation
The article describes a modification to the more commonly known Google PageRank algorithm. The purpose of the approximation is to identify “important” nodes with a focus on scalability, speed, and load balancing amongst a cluster of computational servers. Therefore, this algorithm describes an efficient fully-distributed algorithm for computing PageRank.
In our class, we discussed iterative methods for calculating the PageRank of nodes within a graph, according to Google’s PageRank algorithm. However, due to the need for iterative implementations that require the entire graph’s values to be calculated before each step, with exponentially increasing number of nodes, this method is extremely complex. Therefore, due to this, there are algorithms being created that determine PageRank of nodes based on random walks through the graph. This new method is extremely conducive to computing PageRank of nodes in a distributed computing context. This allows for a company like Google to be able to determine the PageRank of a singular node without necessarily recalculating the PageRank of every node in the graph.