Subject: Re: RAIDFrame One or Many
To: Manuel Bouyer <bouyer@antioche.eu.org>
From: Greg Oster <oster@cs.usask.ca>
List: netbsd-users
Date: 04/15/2003 18:02:16
Manuel Bouyer writes:
> On Mon, Apr 14, 2003 at 02:09:39PM -0700, Caffeinate The World wrote:
> > I have two disks sd0 and sd1. I want to make a RAID1 from both for /,
> > swap, /usr, and /var:
> >
> > Partition Mount
> > a /
> > b swap
> > d /var
> > e /usr
> >
> > Would it be more efficient to make one huge partition of type RAID on
> > sd0 and sd1, then make raid0a, raid0b, raid0d, raid0e or make 4
> > different RAIDs: raid0a, raid1a, raid2a, raid3a with each having a
> > partition like above from sd0 and sd1?
>
> Probably better with only one raid,
I tend to recommend one separate RAID set per filesystem.. After a system
outage (crash/power failure, whatever) smaller RAID sets will rebuild faster
and get into "redundant" operation sooner. As well, if you get a bad block
on one disk, it doesn't affect *all* of the RAID sets, just the one that has
the bad block.
> as RAIDFrame has some I/O scheduling
> knowledge, which are of little use when there are different raid volumes
> on the same physical disks.
I'm not sure whether or not the I/O scheduling in RAIDframe really buys that
much... There are some I/O throttle limits in RAIDframe that may actually
make having multiple RAID sets perform better than just one big one...
Of course, if someone would like to do some emperical benchmarking of this,
that'd be great! :)
> However, because of a problem with swap on raidframe, I would make one raid
> for a+b and a second for everything else (this is how I setup my sustems
> these days). The problem is that the swap partition isn't closed on
> halt/reboot, so the raid is dirty on next boot and it rebuilds parity.
> Using a separate, smaller raid for swap will make the parity rebuild
> short enouth to not be too annoying.
>
> --
> Manuel Bouyer <bouyer@antioche.eu.org>
> NetBSD: 24 ans d'experience feront toujours la difference
> --
Later...
Greg Oster