>From an old IBM RVA Redbook:
"RAID 6 in general is an extension of RAID 5 with a second portion of
parity data to ensure disk recovery in the event of a second disk failure
before the first drive can be fixed."

I believe that the EMC Symmetrix boxes can have more than two disks
involved in a mirrored "pair".

Conceptually it would seem to me that you could come up with an arbitrary
"raid" implementation that provides protection from any number of failures:
What about mirroring your entire RAID5 rank?  Or mirroring a RAID6 rank?
Or mirroring a mirror of a mirror?

Of course the problem is that the overhead in maintaining these complicated
schemes would undoubtedly likely prove more trouble than it's worth.  You
need to look at all contributors to your availability.  Even if you have a
RAID-6 RVA, if you plug both power cords into the same PDU and that PDU
goes down, you're at least going to lose access to your data for a while.
Not that I would have ever heard of anybody doing anything so silly... :)

Scott Chapman




                      David Boyes
                      <dboyes@sinenomin        To:       [EMAIL PROTECTED]
                      e.net>                   cc:
                      Sent by: Linux on        Subject:  Re: raid question
                      390 Port
                      <[EMAIL PROTECTED]
                      IST.EDU>


                      02/10/03 02:47 PM
                      Please respond to
                      Linux on 390 Port






> I was told that if you lose one volume of a raid set you can replace
> and the data will get rebuilt, but if you lose a second volume before
> you get the first one rebuilt you will lose all the data --
> irretrievably!
> Is this true?

Yes, for RAID 5. While the parity information *is* distributed, it's a
one-dimensional distribution. When that second drive goes and the
rebuild isn't yet complete, you're SOL. In RAID 0 (striping only), lose
one and you're screwed.  For RAID 1 (mirroring), they're mirrored pairs,
so if you lose one and then another one, the data is gone and there's
nothing to recover from.

> Is there any reasonable way to get around it?

Y-cabled parallel drive arrays like they use in telcom switches and
nuclear explosion instrumentation, but the price tag is astronomical, so
probably not in the "reasonable" class. 160G of that equipment is well
over $750K.

> Is there
> a reasonable alternative to raid?

AFAIK, no, not unless you happen to be the Sultan of Brunei and have a
immense quantity of unused cash available with nothing better to do with
it (and if you do, I'll be happy to give some of it a warm home...).

- db

Reply via email to