Messages in this thread | | | Date | Fri, 6 Jan 2012 08:33:33 -0800 | From | Tejun Heo <> | Subject | Re: [RFC PATCHSET RESEND] ioblame: statistical IO analyzer |
| |
Hello, again.
On Fri, Jan 06, 2012 at 08:02:35AM -0800, Tejun Heo wrote: > Hmmm... originally, I had variable length data structure there but now > it's fixed so exposing them using tracepoint shouldn't be too > difficult and could actually be better (previously it didn't really > fit TP and ringbuffer should be used directly). Yeah, that's a > thought. Generating a TP event per IO shouldn't be taxing and it > would give much better visibility to userland and we can drop the > whole statics configuration and stuff. Enticing. I'll think more > about it.
The more I think about it, the better it seems. I ended up putting the statistics gathering inside the kernel mostly because the information I had at IO completion wasn't too fit to export to userland verbatim, but while iterating the implementation those parts got chopped (it was necessary for statistics too anyway) and now it seems I can just replace the whole thing with a single tracepoint and drop the whole configurable statistics stuff along with act and its reclaiming, and silly _bin interface. That's just sweeet. Someohw I completely missed the transition of the information available at IO completion.
I might be missing something important but will give it a shot and, if it goes as expected, post updated series soon.
Thanks a lot for bringing it up. :)
-- tejun
| |