Combinatorial BLAS 2.0: Scaling combinatorial algorithms on distributed-memory systems

TL;DR

Combinatorial BLAS 2.0 enhances distributed-memory combinatorial algorithms with features like communication avoidance, GPU support, and scalable I/O, enabling efficient parallel processing for graph analysis and related fields.

Contribution

This paper introduces key technical advancements in Combinatorial BLAS 2.0, including communication avoidance, hierarchical parallelism, GPU acceleration, and scalable I/O for improved distributed combinatorial computations.

Findings

Enhanced scalability and performance in distributed combinatorial algorithms

Effective use of GPU kernels and multithreading for acceleration

Guidelines for selecting data structures and functions in various scenarios

Abstract

Combinatorial algorithms such as those that arise in graph analysis, modeling of discrete systems, bioinformatics, and chemistry, are often hard to parallelize. The Combinatorial BLAS library implements key computational primitives for rapid development of combinatorial algorithms in distributed-memory systems. During the decade since its first introduction, the Combinatorial BLAS library has evolved and expanded significantly. This paper details many of the key technical features of Combinatorial BLAS version 2.0, such as communication avoidance, hierarchical parallelism via in-node multithreading, accelerator support via GPU kernels, generalized semiring support, implementations of key data structures and functions, and scalable distributed I/O operations for human-readable files. Our paper also presents several rules of thumb for choosing the right data structures and functions in Combinatorial BLAS 2.0, under various common application scenarios.