Обсуждение: Daily DBT-3 (DSS) Results on CVS head
FYI, I have results being generated daily against CVS, in addition to dbt2: http://developer.osdl.org/markw/postgrescvs/ I've also added a link to instructions on how to submit patches to test against PostgreSQL on that page, if anyone's interested. Thanks, Mark
"Mark Wong" <markw@osdl.org> writes > > http://developer.osdl.org/markw/postgrescvs/ > The last two jump points looks nice! They are pulled from CVS on 5.30 and 6.1. Where the performance improvements come from? I could remember: * 5.30: avoid unnecessary foregin key check Seems CRC patch is not tested yet (on 6.2). Regards, Qingqing
Hello Mark, Mark Wong wrote: > http://developer.osdl.org/markw/postgrescvs/ This site includes "Profile Report". It's very interesting and useful! Then, I compared result of 5/29 with the result of 5/30. new-order transactions per minute 20050529 1779.41 20050530 2320.41 And I looked Profile Report, Top 20 by Tick. http://khack.osdl.org/stp/302458/profile/Framework_Close-tick.top20 http://khack.osdl.org/stp/302468/profile/Framework_Close-tick.top20 5/29 --- 11611612 default_idle 241908.5833 555980 _spin_unlock_irq 11582.9167 527285 __copy_to_user_ll 4119.4141 481664 __copy_from_user_ll 3763.0000 221535 _spin_unlock_irqrestore 4615.3125 --- 5/30 --- 21246982 default_idle 442645.4583 198495 __copy_to_user_ll 1550.7422 192117 _spin_unlock_irq 4002.4375 183016 __copy_from_user_ll 1429.812570284 _spin_unlock_irqrestore 1464.2500 --- It is understood that the frequency by which "_spin_unlock_irq" is called has decreased. Is there something finding there? -- Junji Teramoto
On Fri, 10 Jun 2005 19:17:58 +0900 Junji TERAMOTO <teramoto.junji@lab.ntt.co.jp> wrote: > Hello Mark, > > Mark Wong wrote: > > http://developer.osdl.org/markw/postgrescvs/ > > This site includes "Profile Report". It's very interesting and useful! > > Then, I compared result of 5/29 with the result of 5/30. > > new-order transactions per minute > 20050529 1779.41 > 20050530 2320.41 > > And I looked Profile Report, Top 20 by Tick. > > http://khack.osdl.org/stp/302458/profile/Framework_Close-tick.top20 > http://khack.osdl.org/stp/302468/profile/Framework_Close-tick.top20 It looks like the 20050530 run might be bogus. If we look at the top of the page here: http://khack.osdl.org/stp/302468/results/0/ The number of rollbacks are way too high. There should only be 1% for New Order only. I should to refer to the specification to see if I should be counting those rolled back transaction. My intuition tells me I've made a mistake in doing so. ;) What you were seeing in the profiles are mostly a side effect of that. The database had crashed so all the transactions were failing right away. We can see that here: http://khack.osdl.org/stp/302468/results/0/db/log Mark