Our research revolves around Loris, a new storage stack that replaces traditional file system and software RAID layers. As with any new piece of software, we have to make sure that its performance is not too horrible. We have used a limited set of benchmarks to evaluate our work so far, but we would like to perform more thorough and diverse testing on Loris and its competitors.
For this project, we would like you to extend our simple benchmarking infrastructure to include many more benchmarks. We estimate that this project will consist of the following parts:
Porting more commonly used storage benchmarking suites to MINIX3. We are thinking of a subset out of FileBench
, SPEC SFS
. The last one is not free, but we can purchase it.
Writing a set of scripts that allow all these benchmarks to be run with pretty much no manual effort involved.
Comparing the performance of Loris, MINIX3's own MFS file system, and the Ext2 file system port, using these benchmarking scripts. We are also interested in the performance differences that result from changing certain settings in each of those file systems (such as the used block size), and having the automated benchmarking infrastructure will really help there.
Taking traces of the operations performed by these benchmarks, and analyzing them in many respects, for example to see which spectrum of file system operations they cover and with which frequency, and to find out how closely they match the real-world workload that they are trying to mimic. Essentially, you will be evaluating the evaluation tools.
If you are interested, come talk to us!