[nfsv4] Comments on draft-faibish-nfsv4-pnfs-block-disk-protection-00

[Jim Rees <[e-mail]>]

Here are my comments on draft-faibish-nfsv4-pnfs-block-disk-protection-00

1. The Introduction seems confusing.  I think it needs a more concise
problem statement.  The statement in section 4.1 is good.

2. Another use case:  A SAN is exporting several LUNS, all of which are pnfs
volumes.  But the MDS the client is talking to only uses a subset of these
volumes; the others are used by a different MDS that the client is not
talking to (does not have mounted).  In this case, even after the client
sends the GETDEVICELIST, it still doesn't have the complete list of all pnfs
volumes.

3. A possible problem.  Is it always the case that the pnfs volumes are
dedicated to pnfs and never used for anything else?  The case I'm thinking
of is where a server is exporting a partition that holds, for example, an
ext4 file system.  The server could mount the partition locally, or it could
export it as a data partition over iscsi for pnfs (probably at different
times).  The same partition would require two different GUIDs.  Or what
about an LVM partition that holds pnfs data blocks?

4. Not stated in the draft, I think most OSes already read the partition
table of each device at the time its attached (at boot for local disks and
at login time for iscsi), so the draft doesn't impose any undue burden.
It's an implementation point but maybe worth mentioning.

5. Sort of glossed over is the requirement that each volume have a GPT.  But
I guess that's ok.