miniVite: A Graph Analytics Benchmarking Tool for Massively Parallel Systems
Abstract: Benchmarking of high performance computing systems can help provide critical insights for efficient design of computing systems and software applications. Although a large number of tools for benchmarking exist, there is a lack of representative benchmarks for the class of irregular computations as exemplified by graph analytics. We therefore propose miniVite as a representative graph analytics benchmark tool to test a variety of distributed-memory systems. Graph clustering, popularly known as community detection, is a prototypical graph operation used in numerous scientific computing and analytics applications. The goal of clustering is to partition a graph into clusters (or communities) such that each cluster consists of vertices that are densely connected within the cluster and sparsely connected to the rest of the graph. Modularity optimization is a popular technique for identifying clusters in a graph. Efficient parallelization of modularity optimization-based algorithms is challenging. One successful approach was conceived in Vite, a distributed-memory implementation of the Louvain method that incorporates several heuristics. We develop miniVite as a representative but simplified variant of Vite, to serve as a prototypical graph analytics benchmarking tool. Similar to many graph algorithms, miniVite is characterized by irregular communication patterns, high communication to computation ratios, and load imbalances among participating processes, thus making it a representative benchmarking tool.
Unlike some graph-based methods such as breadth-first search and betweenness centrality, miniVite represents highly complex computational patterns stressing a variety of system features. This can in turn help provide crucial insight for codesign of future computing systems. We believe that miniVite will serve as a platform for benchmarking systems and design communication primitives that will be applicable to a broad set of irregular computing applications as well as a platform for the design of efficient graph algorithms.
Back to The 9th International Workshop on Performance Modeling, Benchmarking, and Simulation of High-Performance Computer Systems (PMBS18) Archive Listing
Back to Full Workshop Archive Listing