Gunrock: A Programming Model and Implementation for Graph Analytics on Graphics Processing Units

Graphs are ubiquitous data structures that can represent relationships between people (social networks), computers (the Internet), biological and genetic interactions, and elements in unstructured meshes. Many practical problems in social networks, physical simulations, bioinformatics, and other applications can be modeled in their essential form by graphs and solved with appropriate graph primitives. Various types of such graph primitives that compute and exploit properties of particular graphs are collectively known as graph analytics. In the past decade, as graph problems grow larger in scale and become more computationally complex, the research of parallel graph analytics has raised great interest to overcome the computational resource and memory bandwidth limitations of single processors.Modern Graphics Processing Units (GPUs) are high-performance, highly parallel, fully programmable architectures with high memory bandwidth and computing power. 1 Their excellent peak throughput and energy efficiency brings acceleration to regular applications that have extensive data parallelism, regular memory access patterns, and modest synchronizations [41].