Hi Luca, On Thu, Aug 25, 2016 at 3:33 PM, Luca Toscano <[email protected]> wrote: > > It looks really great, I like a lot the overall solution but I haven't > reviewed (and understood) the whole set of changes yet.
Thanks, there is room for improvement still I think, for example we can probably avoid some close wakeups from TO_QUEUE_APPEND() by using the TIMEOUT_FUDGE_FACTOR (likewise in event_get_timer_event() with EVENT_FUDGE_FACTOR, provided timers are sensibly above it). > Other than pure code > review, I would like to know how changes like this one have been > tested/accepted/rejected in the past, since it would be really great not to > loose momentum imho. AFAICT no change has ever been rejected if it: - is enough tested, - improves things (performances here, supported by numbers), - does not break API/ABI (not an issue here). So let's work on testing and see if it helps. I have no hardware these days to do such tests, I can't start before September. In the meantime, I also proposed the patch to PR 60006 (the OP seems to be testing this area), hopefully we'll have some numbers from there already. Possibly others will test it too, and report... The latest patch is currently in [1], but I think I'll push updates in PR 60006 instead, which is more related to performances. Regards, Yann. [1] https://bz.apache.org/bugzilla/show_bug.cgi?id=57399#c5
