I just tried this on my dual opteron test rig and didn't notice a difference in performance with noatime set. What did make a difference was moving from fxp to bge network cards. bge supports checksum offloading where fxp only supports interrupt bundling. Freed up another 20% idle during my test runs. I'm now hitting ~23,000 selects/sec, ~42% idle when running select-key.smack from 3 client hosts. If we establish a suitable test mix and distribute the smack/data files at least we can get consistent test results across systems.
Nick 6-CURRENT as of Jun 27th Dual Opteron 246 Tyan K8SD Thunder Pro 4GB RAM my-huge.cnf with tweaks. On Wed, 29 Jun 2005 11:45:33 +0200 Jeremie Le Hen <[EMAIL PROTECTED]> wrote: > Hi Michael, hi Steve, > > > For me this is as fast as I need my database to be but I can understand > > there is a difference here between FreeBSD and Linux that would make you > > prefer it as the db OS choice. > > Could you try mounting the filesystem where the database lives with > the noatime option, and re-run your tests ? IIRC from previous threads > on this subject, Linux doesn't really honor this while FreeBSD does, > which pulls down the performances. > > Regards, > -- > Jeremie Le Hen > < jeremie at le-hen dot org >< ttz at chchile dot org > > _______________________________________________ > freebsd-performance@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-performance > To unsubscribe, send any mail to "freebsd-performance- > [EMAIL PROTECTED]" _______________________________________________ freebsd-performance@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-performance To unsubscribe, send any mail to "[EMAIL PROTECTED]"