Distributed Memory Parallel Algorithms for Massive Graphs

To analyze massive graphs, parallel algorithms became a necessity. In addition to huge processing time, massive graphs pose another challenge of a large memory requirement. These graphs may not fit in the main memory of a single processing unit, and the algorithms must be able to work on a small part of the graph at a time. To work with such massive graphs, distributed-memory parallel systems can be a suitable platform for many graph problems. This chapter discusses distributed-memory parallel algorithms that are designed to work with massive graphs for several class of graph problems: random graph generation, network connectivity problems, and finding underlying community structure of a network. Specifically, algorithms for the following graph problems have been discussed in this chapter: generating random graphs with Erdos-Renyi and preferential attachment model, switching edges and generating random graphs with given degree sequences, single source shortest path, breadth-first search, counting triangles, and community detection. These algorithms have been developed for three different distributed-memory computation models: message passing interface (MPI), MapReduce-based Hadoop, and Giraph.