Home

GraphBench is a community-driven graph benchmark suite.

Graph and irregular computation is of increasing importance to industry, the government, and the sciences. The core computational component of many “big data”, machine-learning, and security applications is an irregular computation typically done on a graph data structure.

GraphBench is a collection of graph kernels and datasets to aid graph-processing framework creators in developing their systems. We've collected a number of existing datasets and reference implementations across a wide variety of graph analytics platforms. We've also collected Docker images to help get started with the configuration process for some frameworks.

This is a community effort! We need your help to build a representative and relevant benchmark suite. Please join our Google Group, clone our GitHub repo, and contribute your favorite implementations and data sources!

What's inside

GraphBench is not a traditional plug-and-play benchmark suite with a do-everything run script. Instead we provide components you can use to assemble your own evaluation strategy. We believe this is the right strategy given the diversity of graph frameworks out there.

GraphBench has three components:

• Datasets and Synthetic data generators
• Reference implementations
• Instructions on using Frameworks and Docker images

Contributors

GraphBench is hosted at the University of Washington, Department of Computer Science and Engineering, and includes contributions from:

• Oracle

• Hassan Chafi

• Tim Harris

• University of Washington

• Cindy Xin Yi

• Jake Sanders

• Jacob Nelson

• Brandon Holt

• Brandon Myers

• Luis Ceze

• Mark Oskin

• Simon Kahan

This work is generously supported by Oracle Corporation and the National Science Foundation.