0

I am looking for a file system benchmarking tool to test CephFS. This benchmarking tool has to:

  1. Generate a set of files (call it fileset) following a particular distribution, e.g., gamma, empirical, non-archival, or fitted whatever, etc.
  2. Based on the generated fileset, it should simply be able to use that set and run the test, and thus benchmark the performance.
  3. Get the results, done. And re-test after I do some modifications on SSD/HDD algorithms, systems, etc.

I was wondering if you could suggest me any benchmarking tool that generates/receives specific workload (distribution of file sizes). Any other recommendation is kindly appreciated.

Thanks in advance

shoosah
  • 9
  • 1
  • Use 'impressions' to generate the file and directory structure: [home page](http://research.cs.wisc.edu/adsl/Software/Impressions/). ([LWN article](https://lwn.net/Articles/385081/), [publication](http://research.cs.wisc.edu/wind/Publications/impressions-fast09.pdf)) – Davor Cubranic Mar 28 '17 at 23:59

3 Answers3

1

The fio tool is probably your best bet, however, iozone may be able to do what you need as well...

ewwhite
  • 194,921
  • 91
  • 434
  • 799
0

Have a look at Sysbench - fileio mode

HTF
  • 3,050
  • 14
  • 49
  • 78
0

Filebench may do all the things you are looking for.

http://sourceforge.net/projects/filebench/

Noah Watkins
  • 133
  • 7
  • FileBench has couple of bugs: 1. It doesn't accept decimal points (as in probability) 2. It cannot 'reproduce' the fileset (i.e., a set of file with a specific distributions), thus cannot be able to reproduce the experiment! – shoosah Jan 02 '14 at 21:53
  • The distribution parameters are merely scaled: e.g., `dirgamma` value of 100 corresponds to gamma of 0.1. Also for reproducibility, set the `seed` and/or set the `reuse` option to true. – Davor Cubranic Mar 28 '17 at 23:11