SI2-SSE: Gunrock: High-Performance GPU Graph Analytics

点击翻译按钮获取中文摘要

Many sets of data can be represented as "graphs". Graphs express relationships between entities, and those entities and relationships can be used to solve problems of interest in many fields. For instance, a social graph (like Facebook's) links people (entities) by friendships (relationships), and with that graph, Facebook can suggest people to you who might be your friends. Amazon might use a graph made of people and items for sale (entities) connected by who bought those items (relationships) to suggest items you might want to buy. A credit card company might look at your pattern of purchases and detect possible fraud even before you know your credit card was stolen. Graphs are also useful in many fields of science, such as genomics, epidemiology, and economics. This project uses an emerging programmable processor, the graphics processor (GPU), to solve graph problems. GPUs are rapidly moving into our nation's largest data centers and supercomputers. The project team is building a system for computation on graphs that will significantly improve performance on these problems. In this project, the team will work with the computing community and the scientific community, both of whom have numerous interesting, challenging graph computation problems that this system will target. The system is open-source software and can be used freely by researchers and industry all over the world.This project, supported by the Office of Advanced Cyberinfrastructure seeks to develop the "Gunrock" programmable, high-performance, open-source graph analytics library for graphics processors (GPUs) from a working prototype to a robust, sustainable, open-source component of the GPU computing ecosystem. Gunrock's strengths are its programming model and highly optimized implementation. With this work the project team hopes to address Gunrock's usability in the computing and scientific communities by improving Gunrock's scalability, capabilities, core operators, and supported graph computations. In this work the team will collaborate with the GPU Open Analytics Initiative and the NSF-sponsored CINET project for network science to ensure that our work has the broadest possible impact.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.