IRC Logs for #circuits-dev Wednesday, 2013-01-09

prologic>>> (1-(19.6/21.4))*100.000:49
prologic8.41121495327101700:49
prologic>>> 21400-1960000:49
prologic180000:49
prologic>>>00:49
*** Osso has joined #circuits-dev00:51
prologicOsso, see the logs01:07
Ossohello01:07
Ossoyou said "ok" ?01:07
prologicbasically I did what you suggested01:07
prologicrefresh01:08
prologicI need to restart and update circuits.web on that server01:08
prologicI fixed some caching issues with static dispatcher01:08
prologic>>> (1-(19.6/21.4))*100.001:08
prologic8.41121495327101701:08
prologic>>> 21400-1960001:08
prologic180001:08
prologic>>>01:08
prologicbasically this ^^^01:08
prologicI did what you suggested01:08
prologicdid some performance analysis using circuits.bench01:08
prologictrusty ol thing :)01:08
prologicthere is a clear difference of about ~10%01:08
prologicwhich on my Mac at work equates to about ~5000 events/s01:09
prologicand on my AMD Sempron desktop at home here about 1800 events/s01:09
prologicthat is to say01:09
prologiclocking costs about ~10%01:09
prologicor at least the locking mehere implements at revision 327201:09
Osso10% is quite a lot01:10
Ossoit's worth optimizing01:10
Ossobrb01:10
prologicyes I think so01:13
*** prologic has joined #circuits-dev03:21
*** prologic has quit IRC03:22
*** prologic has joined #circuits-dev03:22
mehereWhat precisely did you comare? Revision 3272 and the current state?04:18
mehereMaybe have a look at https://mail.python.org/pipermail/python-ideas/2010-July/007644.html. Having an additional python function invocation in your code costs more than acquiring and releasing a lock! So if you want to compare performance with and without lock, mustn't change anything else.04:25
Ossothat's right locking is more like 5%05:13
Ossoand removing the reduce_time_left call was about 5%05:13
Ossoroughly if they are about the same speed as the benchmark05:14
Ossosays05:14
*** Osso has quit IRC10:16
*** Osso has joined #circuits-dev11:28
*** Osso has quit IRC12:28
prologicmehere, 3272 and 338113:20
prologicmehere, and yes you are quite right13:20
prologicit's not locking necessarily13:20
prologicbtu the way in which we implement it13:20
prologicfunction calls are expensive when you're doing lots of them :)13:20
prologicgotta leave for work, ttl13:20
*** jgiorgi has joined #circuits-dev22:04

Generated by irclog2html.py 2.11.0 by Marius Gedminas - find it at mg.pov.lt!