A scalable, asynchronous spanning tree algorithm on a cluster of SMPs

Large-scale data science applications require manipulating large graphs distributed across multiple processors. In this paper we present our experimental study of an asynchronous, distributed spanning tree algorithm that handles the challenging random, sparse graphs with billions of vertices. With a constant number of barriers, our implementation scales to 1024 processors on a cluster of SMPs. Our algorithm sheds new light on the design and implementation of graph algorithms on distributed-memory machines.