pNFS server completed

[Rick Macklem]

Karli Sjöberg wrote:
[stuff snipped]
>OK, I think I understand, thank you for the explanation!
>
>So you could say that this enhancement is more about the performance
>benifit of scaling out, rather than resiliance/tolerance?
Yes, although without the DS mirroring (which is optional and doesn't have to
be done), you would go from one SPOF to N SPOFs.

To be honest, this pNFS service will not be useful for many. Only people who
need to scale out beyond the limits of one NFS server and find it more convenient
that creating a second NFS server.

Improving resiliance/tolerance for an NFS server is a separate topic (and would
apply to the MDS of this pNFS service as well).
I don't have any expertise w.r.t. this, but something like HA (which I just learned
about from the email thread) might be appropriate. If you can build a storage
subsystem with internal redundancy (raid at the drive level) and then have
two servers "attached" to the storage subsystem so that it can fail over
from the 1st to 2nd server, that would be a good start.
With that, you would just need to have something that would tell the nfs
servers to start/stop when the HA failover happens, I think?
This is a really interesting problem, but not one for me. Of course if someone
gets to the point where it needs an nfsd tweak to make the fail over work, I
could help with that.

Thanks for your interest, rick