[Beowulf] Surviving a double disk failure
billycrook at gmail.com
Fri Apr 10 11:05:29 PDT 2009
On Fri, Apr 10, 2009 at 12:27, Joe Landman
<landman at scalableinformatics.com> wrote:>
> We have 1 customer using 24 drives (22 for data with 2 hot spares) as an md
> raid6 on DeltaV. Normally we'd suggest something smaller (collections of
> RAID6 and then striping across them to form RAID60's).
> With late model kernels, mdadm, and using 1.2 metadata on the md's, you
> should be able to build fairly sizeable stripe width devices. I've heard
> limits of 255, but never tested this far out.
Citing a 2004 paper by H. Peter Anvin entitled 'The mathematics of
RAID-6', Raid-6 (as implemented in the Linux kernel) can support a
maximum of 257 drives. (255 drives worth of data, and 2 drives worth
of parity, distributed evenly of course). It's a limitation of the
Galois Field algebra upon with raid6 is based in Linux.
To determine how many disks you want in a raid array, do some research
on that model's MTBF and sustained throughput; and make sure the
probability of an unrecoverable error happening on any one of the
drives during rebuild is low enough for your comfort. As a very,
very, general rule, you might put no more than 8TB in a raid5, and no
more than 16TB in a raid6, including what's used for parity, and
assuming magnetic, enterprise/raid drives. YMMV, Test all new drives,
keep good backups, etc...
More information about the Beowulf