On the Accuracy and Scalability of Intensive I/O Workload Replay

Alireza Haghdoost and Weiping He, University of Minnesota; Jerry Fredin, NetApp; David H.C. Du, University of Minnesota

15th USENIX Conference on File and Storage Technologies (FAST 2017)
Feb. 27 – March 2, 2017 Santa Clara, CA

We introduce a replay tool that can be used to replay captured I/O workloads for performance evaluation of high-performance storage systems. We study several sources in the stock operating system that introduce the uncertainty of replaying a workload. Based on the remedies of these findings, we design and develop a new replay tool called hfplayer that can more accurately replay intensive block I/O workloads in a similar unscaled environment. However, to replay a given workload trace in a scaled environment, the dependency between I/O requests becomes crucial. Therefore, we propose a heuristic way of speculating I/O dependencies in a block I/O trace. Using the generated dependency graph, hfplayer is capable of replaying the I/O workload in a scaled environment. We evaluate hfplayer with a wide range of workloads using several accuracy metrics and find that it produces better accuracy when compared with two exiting available replay tools.