DFS-Perf Overview

DFS-Perf is a unified benchmarking framework for evaluating the performance of various DFS. DFS-Perf offers good scalability by supporting multi-node, multi-process, and multi-thread testing modes. These modes can work in a combination way in DFS-Perf. Morevoer, DFS-Perf provides typical workloads to test on a series of widely-used underlying file systems, e.g. GlusterFS, GPFS, HDFS and Alluxio. And DFS-Perf is quite extensible to easily plugin user-defined workloads and support new DFS.

For the users of DFS, DFS-Perf is an evaluation utility to choose the most suitable DFS for applications. And for the developers of DFS, DFS-Perf is a good utility optimize and improve DFS itself.

Gitbucket Repository | Releases and Downloads | User Documentation | JIRA | User Mailing List

Current Features

User Documentation

Running DFS-Perf Locally: Get DFS-Perf up and running on a single node for a quick spin in ~ 2 minutes.

Running DFS-Perf on a Cluster: Get DFS-Perf up and running on your own cluster.

Running DFS-Perf on GlusterFS: Get DFS-Perf up and running on GlusterFS.

Running DFS-Perf on GPFS: Get DFS-Perf up and running on GPFS.

Running DFS-Perf on HDFS: Get DFS-Perf up and running on HDFS.

Running DFS-Perf on Alluxio: Get DFS-Perf up and running on Alluxio.

Workloads: A brief introduction of those workloads in DFS-Perf.

Configuration Settings: How to configure DFS-Perf.

Perf Report: A performance report of a testing cluster.

Support or Contact

If you are interested in trying out DFS-Perf in your cluster, please contact Rong Gu and Qianhao Dong. Users are welcome to join our mailing list to discuss questions and make suggestions. We use JIRA to track development and issues.

Acknowledgement

DFS-Perf is a research project started at the Nanjing University PASA Lab and currently led by Rong Gu & Qianhao Dong. We would also like to thank to our initial project contributors in PASALab.