Lustre discuss - Apr 2009 - Network raid6?

For redundancy why not a raid like parity as it stripes across the
ost''s. For example 100 oss''s with directly attached raid6
arrays. then
have 2 parity stripe blocks over the 100 allowing for 2 failures of
oss''s or the directly attached ost''s. This seems more cost
effective
than mirroring using drdb or heartbeat. In fact it eliminates the need
for either.

On Apr 09, 2009  03:48 -0400, Christopher Deneen wrote:
> For redundancy why not a raid like parity as it stripes across the
> ost''s. For example 100 oss''s with directly attached raid6
arrays. then
> have 2 parity stripe blocks over the 100 allowing for 2 failures of
> oss''s or the directly attached ost''s. This seems more
cost effective
> than mirroring using drdb or heartbeat. In fact it eliminates the need
> for either.
Sure we''ve thought of this, but implementation is MUCH more difficult
than describing it, believe us.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.