Subject: Re: RAIDFrame in Production Use
To: Simon Burge <simonb@wasabisystems.com>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: netbsd-users
Date: 04/12/2003 01:48:15
On Sat, Apr 12, 2003 at 03:44:20PM +1000, Simon Burge wrote:
> Andrew Gillham wrote:
> 
> > On Fri, Apr 11, 2003 at 01:54:56PM +0200, Christoph Kaegi wrote:
> > > On 2003.04.10 16:23, Caffeinate The World wrote:
> > > > If you have experience in using RAIDFrame in production, I'd like to
> > > > know what your thoughts are. 
> > > 
> > > All the servers I run (about 10) are completely mirrored
> > > with RAIDFrame (/, /usr, /var, /home, swap )
> > > 
> > > I've never had troubles with it.
> > 
> > Are people typically using one raid volume for all file systems or a 
> > raid volume for reach file system?
> 
> I'm using 4 raid1 sets with /, /var, /data and swap.
> 
> The main reason I started with separate sets is that swapping to a raid
> set results in problems where the raid set wasn't shut down properly on
> reboot so it needs it parity rebuilt with each reboot.  From there is
> seemed obvious to use separate sets for each filesystem too.

I think this is actually a pretty bad idea; it will defeat both the
seek optimization in RAIDframe *and* the seek optimization in the disk
drivers.  Running multiple filesystems on a single physical spindle is
not the greatest thing in the world, from a performance point of view;
this, it seems to me, will make it significantly worse.

Thor