which means that shared_buffers can be increased more without performence loss
2006-06-07 15835, 2006
nikki_
gaaah. I have maths coming out of my ears.
2006-06-07 15845, 2006
catgroove
dakar nikki1
2006-06-07 15847, 2006
Shepard
you already added some instruments for me. you should have said that earlier :)
2006-06-07 15852, 2006
catgroove
damn
2006-06-07 15800, 2006
catgroove
ok send me as payement then
2006-06-07 15813, 2006
catgroove
that was sorta an intended typo
2006-06-07 15824, 2006
catgroove
dunno wtf a paye is though
2006-06-07 15848, 2006
ruaok reads
2006-06-07 15832, 2006
Russss
ruaok: also you should look into pgpool if you're not using it - it's a fairly shitty bit of software but it does the job.
2006-06-07 15815, 2006
Russss
although I don't know how much pooling perl does. Maybe it's not that necessary
2006-06-07 15815, 2006
ruaok
how does that compare to DBI ?
2006-06-07 15828, 2006
ruaok
we're already reusing connections via DBI.
2006-06-07 15846, 2006
Russss
in retrospect, there's probably very little difference with few web servers using their own pooling
2006-06-07 15813, 2006
Russss
when you have 15 web servers it's inefficient to have each of them keep a pool of 20 connections each. So centralised pooling is better.
2006-06-07 15822, 2006
ruaok
I grok that and will keep that in mind.
2006-06-07 15828, 2006
ruaok
we still have only one web server machine.
2006-06-07 15837, 2006
Russss
yeah so no difference
2006-06-07 15839, 2006
Russss
ok
2006-06-07 15841, 2006
catgroove
I dig the word grok
2006-06-07 15807, 2006
BrianG
i grok the word dig
2006-06-07 15823, 2006
ruaok
Russss: have you played with tweaking the work_mem setting?
2006-06-07 15849, 2006
Russss
yes, it's very application-specific though
2006-06-07 15806, 2006
Russss
if you know you're only doing simple sorting you can crank it up quite a bit
2006-06-07 15814, 2006
Russss
but beware it's per-sort, per-query
2006-06-07 15834, 2006
ruaok
under what conditions do I not want to crank it up.
2006-06-07 15841, 2006
ruaok
what value are you using?
2006-06-07 15849, 2006
catgroove is away: catgroove macht die Wäsche
2006-06-07 15818, 2006
Russss
when you have n queries doing a sort, it'll use n*work_mem
2006-06-07 15825, 2006
Russss
we have it at 8192
2006-06-07 15851, 2006
Russss
maybe that's slightly conservative for a 16GB machine though
2006-06-07 15858, 2006
ruaok
LOL.
2006-06-07 15806, 2006
ruaok
we're half that for a 4GB machine.
2006-06-07 15818, 2006
ruaok
next time things go to shit, I will try upping that a little.
2006-06-07 15828, 2006
ruaok covets Russss 16GB machine
2006-06-07 15844, 2006
Russss
I would recommend looking into more memcaching though
2006-06-07 15817, 2006
Russss
that's how we can do 200 requests per second on a single db server
2006-06-07 15830, 2006
keschte joined the channel
2006-06-07 15845, 2006
Russss
actually it's closer to 300 if you count submissions
2006-06-07 15827, 2006
SoothingR has quit
2006-06-07 15832, 2006
dsp
last.fm?
2006-06-07 15839, 2006
Russss
yes
2006-06-07 15858, 2006
dsp
nice service you have :)
2006-06-07 15802, 2006
ruaok
Russss: that is certainly the case.
2006-06-07 15852, 2006
Russss
check the hit rate on your memcache servers, if it's lower than 95%, get more RAM :)
2006-06-07 15816, 2006
ruaok
how do you check that?
2006-06-07 15831, 2006
ruaok
our mrtg graph for memcached isn't hooked up yet.
2006-06-07 15849, 2006
Russss
telnet into the memcache server and type stats
2006-06-07 15806, 2006
ruaok
and, I know we need to do more caching and there are dozen of places I could think of. but when it comes to opimization, I have a shitty instint. I want hard numbers before I write code.
2006-06-07 15810, 2006
Russss
it gives you a load of stats including hit ratio
2006-06-07 15816, 2006
ruaok
cool, thanks.
2006-06-07 15831, 2006
ruaok
how do you suggest we go about quantifying what things need caching the most?
