Subject: Re: Raid-0
To: None <admin@raid.netbsd.org>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: port-i386
Date: 08/05/2001 15:03:24
On Sun, Aug 05, 2001 at 11:09:08AM +1000, admin@raid.netbsd.org wrote:
> On Saturday 04 August 2001 11:39,  Manuel Bouyer wrote:
> >
> > The main downside is performances problems: with stripping, data are
> > equally distributed across the component ("interleaved"), so when you read
> > a large file data are fetched from all component providing a performances
> > boost. Concatening will just put component one after the other (sector 0 to
> > x are on component 0, x to y on component 1, etc).
> > To see the difference, do a 'dd' from your raid0 raw device to /dev/null.
> > With stripping, you'll see activity on all component. With concatenation
> > you'll see activity on the first component, then the second, etc ...
> >
> 
> Everytime I shutdown the raid-10 server, the parity of the raid sets have to 
> be reconstructed. Is this to do with the fact that my raid-0 disk sets are 
> not the same size, ie one 20G and 15G?

No. You have several raid-1 devices concatened with a raid-0, rigth ?
What happens is that the raid-0 doesn't close its devices at shutdown, so the
underlying devices can't set the partity to clean (as they have devices
open). I think this is now fixed in -current.
As a workaround you can setup a rc.shutdown script which will unmount
your raid-0 and unconfigure it (obvisously this can't work if / is
on this raid volume).

--
Manuel Bouyer <bouyer@antioche.eu.org>
--