Date: Sat, 01 Apr 2006 12:30:46 +1200 From: Mark Kirkwood <markir@paradise.net.nz> To: Eric <ericd@free.fr> Cc: Lucas Holt <luke@foolishgames.com>, freebsd-smp@freebsd.org Subject: Re: Low perf with smp Message-ID: <442DC9B6.2090200@paradise.net.nz> In-Reply-To: <01a901c65493$77ad2740$65fd24c0@Eric> References: <009301c65418$cc029f30$65fd24c0@Eric> <64659004-E643-4044-AAED-076E0CC91977@foolishgames.com> <01a901c65493$77ad2740$65fd24c0@Eric>
next in thread | previous in thread | raw e-mail | index | archive | help
Eric wrote: > Thanks a lot Lucas. > What's sound so strange to me is that there we so much difference > 30sec=> 54 sec simply by activating smp in kernel. > But I will search on this way and see if it don't come from mysql. > > I will send you issue. > > Eric. > ISTR there was some business with Mysql calling gettimeofday a lot, and therefore being particularly impacted by the sysctl 'kern.timecounter.hardware'. As I understand it, when you activate SMP a more accurate, but slower timecounter is chosen (typically 'i8254' instead of 'TSC' on intel HW). It might be interesting to see what happens to your times by forcing TSC and re-running the test (ahem - better only do this on a test box, as there is bound to be a good reason for the more accurate timecounter being chosen in SMP mode....). Cheers Mark
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?442DC9B6.2090200>