On Tue, 2010-10-26 at 09:29 +1300, Ian Collins wrote:
> On 10/26/10 08:53 AM, Edward Martinez wrote:
> >> ZFS can not do it's magic with
> >> error detection and correction if you use hw raid.
> >>      
> > Thanks!
> >
> > I was not aware ZFS loses many of its features when used  with hw RAID 
> > cards. In that case I'm  not getting it
> >
> >    
> It doesn't loose many (or any if you just serve JBOD from the RAID 
> card), just one of the most important - control of redundancy!
> 


Well...

As always, the issue is more complex.  :-)

Realistically, you want ZFS to handle all of the RAID configuration, to
make maximum use of its features.  That means, for the most part, you
will bypass all the extra features of a Hardware Raid Controller, likely
running it in either "JBOD Mode", or simply configuring it to export
each individual disk as a Raid 0 LUN.

However, there are still a couple of major features in a HW controller
that you might be interested in:

(a) for boot drives, ZFS (like all software RAID systems) is UNABLE to
automatically (i.e. unattended) boot from the secondary mirror half if
the primary boot drive fails.  This is a BIOS limitation - if you use
any software raid solution, and disk 0 fails, you will have to go to the
BIOS itself to instruct it to boot from disk 1.   (I'm slightly
over-simplifying, but this is true for the vast majority of systems and
HBA/Raid cards).  If you use a RAID card and create a mirror, it
presents this to the BIOS as a single disk, and thus, you can reboot if
either dies.

(b) ZFS currently supports only single vdev rpool pools, and the vdev
must either be a mirror or single disk.  HW controllers have no such
restrictions. 

(c) If (and ONLY IF) you have a battery-backed NVRAM cache on the Raid
Controller, this can provide a significant performance boost over a
plain HBA.  This cache can be used even if you don't use the HW Raid
functionality.  If you have a cache, but no battery, then I'd really not
recommend using that kind of Raid Controller at all. Data loss looms.


-- 
Erik Trimble
Java System Support
Mailstop:  usca22-317
Phone:  x67195
Santa Clara, CA
Timezone: US/Pacific (GMT-0800)

_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org

Reply via email to