On Mar 7, 2007, at 9:27 AM, Robert Thurlow wrote:

> I propose a new project that delivers the NFSv4 extensions for
> Mirror Mounts, Referrals, and Replication & Migration.  This
> project should be endorsed by the NFS community.
>
> What's a mirror mount?  It's a way to make the client automatically
> mount filesystems from the same server as it discovers them.  The
> current Solaris NFS client code does not do this, which forces
> admins to update their automounter maps when new filesystems are
> created.  ZFS admins would rather be free to create new filesystems
> and have them discovered.  Mirror mounts are described in the NFSv4
> RFC 3530, related to the server namespace.  For more about how we
> would like this to work, please see:
> http://blogs.sun.com/tdh/entry/how_nfsv4_should_work_when
>
> What's a referral?  It's a pointer a server can give to a client
> that means, "I don't have the filesystem, but you can find it
> here."  The client will redirect its access to the referred
> location transparently and continue.  A global namespace can be
> constructed with a mechanism to find an initial global root and
> referrals.  The current project is working on NFSv4.0 (RFC 3530)
> referral support on the client, with work to follow on NFSv4.1
> referral extensions.
>
> What's replication & migration?  RFC 3530 defines support for a
> server to be able to say, "that filesystem is no longer here,
> it's over there now" or "another copy of this filesystem lives
> here."  This links the idea of a referral to the actions needed
> to resynchronize state of files the client was already using prior
> to the migration event.  Being able to move and replicate filesystems
> freely is another part of a global namespace.

This work should be done and as suggested, will be endorsed by
the NFS community.

With respect to scope, does this proposal cover client and server
and if server, how much support or level of integration is
anticipated for things like referrals and replication/migration?
Is there an expectation of loose or tight integration with
underlying filesystems like ZFS?

Spencer

Reply via email to