Efficient Compression on Real World Directed Graphs

Real world graphs typically exhibit power law degrees, and many of them are directed graphs. The growing scale of such graphs has made efficient execution of graph computation very challenging. Reducing graph size to fit in memory, for example by using the technique of lossless compression, is crucial in cutting the cost of large scale graph computation. Unfortunately, literature work on graph compression still suffers from issues including low compression ration and high decompression overhead. To address the above issue, in this paper, we propose a novel compression approach. The basic idea of our graph compression is to first cluster graph adjacency matrix via graph structure information, and then represent the clustered matrix by lists of encoded numbers. Our extensive evaluation on real data sets validates the tradeoff between space cost saving and graph computation time, and verifies the advantages of our work over state of art SlashRarn [ 1, 2] and Ligra+ [3].