[Lustre-discuss] Parity Declustering and Distributed Sparing

[Brian J. Murrell]

On Mon, 2008-10-06 at 05:43 -0500, Daniel Ferber wrote:
> A customer and I were talking about Parity Declustering and Distributed
> Sparing. We don't really understand the presence of either options in the
> Lustre system.

I have not heard of either of these in the context of Lustre before.  Do
you have some documentation relating them?

> It has been our understanding that past instantiations of
> Lustre rely entirely on the enterprise storage beneath the Lustre system in
> order to achieve High Availability.

Indeed.  Lustre adds no amount of redundancy to the storage it's running
on.  The storage system must be able to deal with device failures
itself.  Lustre is only as resilient as the storage it's running on.

> I.e., if one were to want RAID-6
> functionality from a local Lustre node, the one would install Lustre on a
> storage device running RAID-6 in hardware,

Indeed.  Exactly.

> and if one were to want
> split-path failover, you'd buy (for example) two storage nodes, and rely on
> the storage for the failover. What are we not understanding?

I don't think you are misunderstanding anything.  I wonder what the
source of the discussion about "Parity Declustering and Distributed
Sparing" was.

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20081006/1a674209/attachment.pgp>