On Sun, Mar 24, 2013 at 9:04 PM, Justin Clift <jclift@xxxxxxxxxx> wrote: > On 24/03/2013, at 8:55 AM, Stephan von Krawczynski wrote: > <snip> >> ... the most >> important question: the performance tuning. We all know that glusterfs can be >> unbelievably slow if you tuned the wrong switches. And really, finding the >> right ones is very hard, even for very experienced administrators. >> So instead of becoming "autonomous" (i.e. never be successful, I mean you just >> said you cannot solve this by yourself and want to drop it on external >> know-how) it would be a lot better to make some tuning-tool that analyses a >> situation and gives hints how performance can be optimized - but not tuning it >> by itself of course. > > Hmmm, some form of situational tuning sounds like an interesting idea. > > Any idea what kind of statistics/counters it would need to be aware of or > somehow be able to measure? > > Regards and best wishes, > > Justin Clift > > -- > Open Source and Standards @ Red Hat > > twitter.com/realjustinclift > Simplest case is, we supply different profiles. Using gluster cli, admin should be able to set these volume options by simply selecting a profile as a volume property. We collaboratively collect these metrics from Red Hat performance team and the community. To begin with, this can even be a shell script over gluster cli. Ideally I would like to see a tool that generates load on a gluster volume with various kinds of work loads (or user chosen one) simultaneously trying out different volume options with a range of known good values. In the end, it makes recommendations. With that we should also have gluster save-profile and restore-profile commands. Luckily, there are only few volume options that impact performance significantly. So this is doable. -ab Imagination is more important than knowledge --Albert Einstein
