Обсуждение: PostgreSQL data dir on linux raid software
Hi, after some test i got that my performance problem described in the message http://archives.postgresql.org/pgsql-admin/2007-01/msg00116.php is issued by the linux software raid 1 on my discs (yes the silicon image controller has only fake raid hardware). I found this old message that seems the same problem like mine. http://archives.postgresql.org/pgsql-admin/2003-12/msg00219.php Because the message is quite old I'm wondering if there're some hdparm settings to enable on the disks that compose the array to have better performances in fsync. For now I cannot mount an expensive scsi raid hardware so i need to have raid software. Thanks in advance Best Regards Federico Campoli
On Mon, Jan 15, 2007 at 18:24:19 +0100, Federico <rotellaro@gmail.com> wrote: > Hi, > after some test i got that my performance problem described in the > message http://archives.postgresql.org/pgsql-admin/2007-01/msg00116.php > is issued by the linux software raid 1 on my discs (yes the silicon > image controller has only fake raid hardware). > > I found this old message that seems the same problem like mine. > > http://archives.postgresql.org/pgsql-admin/2003-12/msg00219.php > > Because the message is quite old I'm wondering if there're some hdparm > settings to enable on the disks that compose the array to have better > performances in fsync. > > For now I cannot mount an expensive scsi raid hardware so i need to > have raid software. Are you using Linux's software raid (md devices) or dmraid using the controller to do some of the work? If you aren't using the native linux software raid, you might want to try it instead. You will need to rebuild the array to do this, so this might not be an easy thing to do. The bios will also need to support disabling the raid and exposing the individual disks (but that is a fairly normal thing for bios's). If linux can see the individual disks, then you should be able to use hdparm to tweak their settings.
On 1/16/07, Bruno Wolff III <bruno@wolff.to> wrote: > On Mon, Jan 15, 2007 at 18:24:19 +0100, > Federico <rotellaro@gmail.com> wrote: > > Hi, > > after some test i got that my performance problem described in the > > message http://archives.postgresql.org/pgsql-admin/2007-01/msg00116.php > > is issued by the linux software raid 1 on my discs (yes the silicon > > image controller has only fake raid hardware). > > > > I found this old message that seems the same problem like mine. > > > > http://archives.postgresql.org/pgsql-admin/2003-12/msg00219.php > > > > Because the message is quite old I'm wondering if there're some hdparm > > settings to enable on the disks that compose the array to have better > > performances in fsync. > > > > For now I cannot mount an expensive scsi raid hardware so i need to > > have raid software. > > Are you using Linux's software raid (md devices) or dmraid using the controller > to do some of the work? If you aren't using the native linux software raid, > you might want to try it instead. You will need to rebuild the array to do > this, so this might not be an easy thing to do. The bios will also need to > support disabling the raid and exposing the individual disks (but that is > a fairly normal thing for bios's). If linux can see the individual disks, > then you should be able to use hdparm to tweak their settings. > Hi Bruno, thank you for the answer. Yesterday I resolved the issue. The problem cause was the APIC functionality on my motherboard that messed the flush on the disk. Disabling the apic in the bios I got a good performance increase. With ab i made a test with 1000 request on concurrency 400 and got 20 seconds of mean response time without instance tuning. :D Best Regards Federico Campoli