[Gllug] NFS problem

John Hearns john.hearns at streamline-computing.com
Mon Oct 22 19:46:26 UTC 2007


On Mon, 2007-10-22 at 20:34 +0100, John Hearns wrote:
> O
> Sadly, in the era of Tbyte drives, the probability of dual failures
> during reconstruction of RAID-5 sets is getting significant
> 
> http://www.eweek.com/article2/0%2C1895%2C2168821%2C00.asp
> 
> (see the answer to the last question)

I should make my reply a bit clearer here.
Disks have increased hugely in capacity from the days when the original
RAID concept was implemented. Let's say from 40gig disks to the 1Tbyte
disks which are now common.
However, the error rate stays the same (OK, I'll probably get some hard
facts in reply to this, but it is broadly true).
Reconstruction times for a failed RAID-5 set are longer with larger
disks, and during that time you are more likely to hit a disk error,
resulting in tears and storage engineers being sent to bed with no tea.


At the risk of mentioning Panasas again (the scars from the last one are
just healing), this is an advantage of object-based filesystems. Rather
than your entire RAID-5 set being toast if there is a double error, only
that one file which had a part on that disk is fenced off and made
unavailable.


And of course I know that RAID-6 exists, and counters the double
failure. I've implemented RAID-6 storage for several purposes.
And yes, it is possible to have a mirrored RAID5 setup.
And yes, RAID(N) is not backup.





-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list