About power and performance

Power and Performance matter. But Power and Performance are sneaky; if you do not constantly measure either, something as complex as a Linux operating system is guaranteed to slowly but steadily regress. Emperically, by at least 10% per year, but of couse, one can argue long and hard about that rate.
For Fenrus Linux, rather than arguing, we've decided to just meaure every build, every release we do, all the time. And because measurement in a vacuum is useless, we tweaked our reporting to be against 'best of industry'. This means we run the same benchmarks on the same machine on several different distros, and for each test, we pick the best result from those runs, and compare ourselves against that.
Not because we're competing againt those distros on performance, but because we know that the result of this process is at least possible and proven.
Now, currently the list of 'best of industry' isn't very large; it takes a long time to run each test enough times to get rid of the noise that we've only had time so far to do a few. But we'll add more over time... the principle does not change.

The tests we use come from the Phoronix Test Suite. It's easy to run PTS in an automated setup, and it's easy to use it to get repeatable numbers from it.
Over time, if we feel PTS is lacking testing in area we care about, we'll be adding tests to the upstream PTS project to fill any such gaps as they arise.


For all graphs, the results are index such that a higher number always means a better result.



FPU Compute


Media encode/decode

OS overhead