On Thu, Aug 25, 2005 at 03:48:18PM +0100, Tony Finch said:
> On Thu, 25 Aug 2005, Walt Reed wrote:
> > On Thu, Aug 25, 2005 at 02:22:16PM +0100, Tony Finch said:
> > > On Thu, 25 Aug 2005, Walt Reed wrote:
> > >
> > > > In addition to other suggestions that were put forth, making sure your
> > > > RAID controller has battery backed write cache does AMAZING things for
> > > > performance.
> > >
> > > And check the disk write caches are turned off :-)
> >
> > How so? I found the write cache sped most things up a LOT... Found it
> > helped my database by a factor of 2. Added the cache on all my machines
> > and overall load levels have dropped (as reported by MRTG. Especially
> > the IOWait state.)
>
> If you have a battery-backed write cache on the controller, you don't need
> or want a volatile write cache on the disks. The additional cache on the
> disks will not improve performance and will reduce reliability. If the
> disks lie to the controller about when data has hit stable storage then
> the battery will not protect you from losing data in the event of a power
> outage.
>
> http://www.livejournal.com/community/lj_dev/670215.html
> http://www.livejournal.com/users/brad/2116715.html
Ahh. This is where the confusion was coming in. I was not talking about
the cache on the physical disk. I was talking about the cache on the
RAID controller. I was assuming that you were talking about the cache on
the controller. Yes, disabling write cache on the disk itself IS a good
thing (I just assumed that this was being done.)
