The performance evaluation of Personal Clouds (e.g., Dropbox, Box) is challenging due to their distributed architecture and the absence of realistic workload generators that reproduce the disparate types of user behavior in the system. To solve this, we propose a novel storage benchmarking methodology in which the workload is generated by emulating groups of users that exhibit a similar behavior, namely, user stereotypes. We instantiate this methodology in BenchBox, a distributed performance analysis tool that integrates:
- A complete model that reproduces the aspects of user behavior that matter to system’s performance, including file sizes and contents, locality and burstiness properties.
- A centralized management and real-time monitoring system of Personal Cloud desktop clients.
- BenchBox also uses compact files to represent user stereotypes, enabling researchers to share only the behavior of users, instead of private traces.
The code is available at: https://github.com/cloudspaces/BenchBox