NetStorage: A synchronized trace-driven replayer for network-storage system evaluation

Bingzhe Li, Hao Wen, Farnaz Toussi, Clark Anderson, Bernard A. King-Smith, David J Lilja, David H Du

Research output: Contribution to journalArticle

Abstract

Trace-driven replayers provide a flexible and convenient way to evaluate or debug target systems. Previous researchers focused on the individual impact of either the network or the storage system on overall system performance by using trace-driven replayers. However, as big data and cloud servers become prevalent and the performance gap between storage and networks decreases substantially, it becomes important to consider both impacts of storage and network on target systems. In this paper, we build a replayer called NetStorage to replay the captured traces of both network and storage to evaluate the performance of combined network-storage systems. When involving multiple replayers, the synchronization between these replayers becomes crucial. We introduce a manager component to control the synchronization between replayers. By doing that, the NetStorage is capable of efficiently replaying multiple traces synchronously in order to mimic the real application behavior in network and storage environments for different scenarios. Finally, a case study is provided to compare performance of two systems by using the NetStorage.

Original languageEnglish (US)
Pages (from-to)86-100
Number of pages15
JournalPerformance Evaluation
Volume130
DOIs
StatePublished - Apr 2019

    Fingerprint

Keywords

  • Network-storage system
  • Performance evaluation
  • Trace-driven replayer

Cite this