A Scalable Distributed Parallel Breadth-First Search Algorithm on BlueGene/L Session: System Performance Evaluation Event Type: Paper, Gordon Bell Finalist Time: 10:30am - 11:00am Session Chair: Jeffrey Vetter Speaker(s): Andy Yoo, Edmond Chow, Keith Henderson, William McLendon, Bruce Hendrickson, Umit Catalyurek Location: 602-604 Abstract: Many emerging large-scale data science applications require searching large graphs distributed across multiple memories and processors. This paper presents a distributed breadth-first search (BFS) scheme that scales for random graphs with up to three billion vertices and 30 billion edges. Scalability was tested on IBM BlueGene/L with 32,768 nodes at the Lawrence Livermore National Laboratory. Scalability was obtained through a series of optimizations, in particular, those that ensure scalable use of memory. We use 2D (edge) partitioning of the graph instead of conventional 1D (vertex) partitioning to reduce communication overhead. For Poisson random graphs, we show that the expected size of the messages is scalable for both 2D and 1D partitionings. Finally, we have developed efficient collective communication functions for the 3D torus architecture of BlueGene/L that also take advantage of the structure in the problem. The performance and characteristics of the algorithm are measured and reported. This paper can be found in the ACM and IEEE Digital Libaries Click here for ACM Click here for IEEE Chair/Speaker Details: Jeffrey Vetter (Chair) Oak Ridge National Lab Andy Yoo Lawrence Livermore National Laboratory Edmond Chow DE Shaw Research and Development Keith Henderson Lawrence Livermore National Laboratory William McLendon Sandia National Laboratories Bruce Hendrickson Sandia National Laboratories Umit Catalyurek Ohio State University