Workload-Specific File System Benchmarks

Keith Arnold Smith, Ph.D. Thesis, Harvard University, January 2001.

A fundamental problem with the current generation of file system benchmarks is that they fail to take into account the fact that a file system's performance can vary depending on the workload running on it. Many benchmarks attempt to reduce file system performance to a single number, producing a simple one-dimensional ordering of the systems being tested. Although this may be useful for marketing literature, the performance of file systems in the real world is more complicated. Different workloads place different demands on the file system, and can result in different behavior from the underlying system. A file system that provides superior performance for a web server may have inferior performance when running a software development workload.

HBench-FS addresses this problem by providing a framework for workload-specific file system benchmarks. HBench-FS profiles a file system and a workload separately, and provides a set of tools for combining these profiles to predict the performance of the workload on the target file system. HBench-FS provides detailed predictions of the performance of different parts of the file system interface, allowing researchers and developers to isolate the areas of file system performance that present the greatest bottlenecks for particular workloads.

Full Thesis (159 pages): PDF or gzipped Postscript