The Clustering and Userland VFS transport protocol - summary

[Bill Huey (hui)]

On Sun, May 14, 2006 at 10:36:39AM -0700, Matthew Dillon wrote:
. .. 
>     The initial commit will just replace the VFS and VOP calls with the new
>     infrastructure.  There is actually no change at all to the calls
>     themselves (yet).  It will virtually be a drop-in replacement.  After
>     that I'll work on extending the protocol to add support for
>     over-the-wire transactions for userland VFS support, which will also
>     necessitate changes to the way I/O is handled (particularly UIO's --
>     data transfers to and from userland).
> 
>     It turns out that the basic communications issues involved with writing
>     a userland VFS are exactly the same issues involved with implementing
>     a communications link between machines in a cluster.

I've been out of this discussion so I apologize if I'm covering material
that's been discussed already. Have you considered, in your design, file
systems like XFS in how they can use seperate journaling devices within
your VFS userland work ? XFS specifically has a function that moves the
journaling state forward and the VFS layer typically is agnostic to
those specific journaling internals. Just curious, how are you going to
resolve that design mismatch ?

Have a Happy Mother's Day, BTW. :)

bill