On Wed, Apr 23, 2003 at 01:35:20PM +0300, Jukka Marin wrote:
> On Fri, Apr 11, 2003 at 01:08:49PM +0200, Johnny Billquist wrote:
> > On Fri, 11 Apr 2003, Jukka Marin wrote:
> > > The biggest surprise was that RAID1 is slower than the disks made me
> > > expect.  When the RAID is 100% busy, the disks are not, neither is the
> > > CPU.
> > 
> > Isn't that to be expected?
> > You are distributing disk I/O through one device to several other
> > devices. When that device is 100% busy, the underlying disks ought to be
> > (1/n)*100 % busy. Thus, if you only have one disk, it should be 100% busy,
> > but if you have two disks, they should only be 50% busy, both of them.
> 
> Hmm.  If I write to RAID1, doesn't it write to both disks simultaneously,
> making both disks 100% busy when RAID1 is 100% busy?

Yes

> How about reads, does
> it read the same data from both disks, compare the data and then pass it to
> user process or does it read only one disk (doesn't seem so) or does it
> read part of the data from disk 1 and part from disk 2?  If it uses both
> disks, then why is RAID1 50% slower than a single disk?

It should be able to use both, but I suspect a sequencial read may come
from only one disk, depending on your parameters.
Writes should be almost as fast as a single disk, reads should be at last as
fast as a single disk. Maybe you got your raid parameters wrong ?

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 24 ans d'experience feront toujours la difference
--