wiki:FsOpBench

FsOpBench - Filesystem Operations Benchmark

FsOpBench is a file system benchmark targeted at testing interactive performance of busy file systems. fsopbench has three primary functions:

  • Create a artificial file tree with a realistic distribution of file sizes for user home-directories as described in  UV 2005 to provide a reproducible test setup.
  • Tree traversal while measuring the time required for reading the directory, stating the file, reading the first byte of the file and finally the performance for reading the first 128k of the file.

When FsOpBench is started for the first time, it will setup a 20 GB file tree for testing. Make sure that you have about twice as much free space for it to work comfortably. In normal operation, FsOpBench will run three tests.

  • one reader running
  • three readers running
  • three readers and three writers running

The 20GB file system tree is structured so that each reader gets its own part of the tree and as such can work on the disk and not from the cache.

The writers use the same code as was used to generate the original file system tree.

When the program terminates, the output of the writers will be removed.

FsOpBench is written in Perl and should run on all UNIX like systems.

Tobi Oetiker mailto:tobi@…