> That'd be too bad, since CODA is kinda neat.

So is AFS and DCE/DFS.

AFS, DCE/DFS, and CODA should (is) be encrypted (I can only speek for AFS).
Please keep it that way, I don't want to go back to NFS-level security
(flambait).

They all have have the diea of idea of optimisti caching. The server tells
the client with a callback (a rpc function) when a file changes.

They all talk over UDP.

Callbacks (the promise from the fileserver to notify the client when the
file changes) are valid to up to one hour.

AFS have what you called "what"/"where" support, both the client and the
server have a afsUUID that uniqly identifyles the client/server.  But it
isn't passed in the back rpc call.

I don't want to talk throu a proxy (your "single exposed CODA talker")
since that compromises security.

Servers and client poll each other ever 5 minuts.

Now, I can't really get that working other with NAT/PAT.

As I see it there are two solution. Use tcp (since most NAT/PAT keep that
state longer then UDP) and/or poll the server ever 7 second (A cisco [67]00
ISDN router with PAT keep UDP state for 15 seconds).

There need to be a lot of statekeeping in the NAT, and since the polling,
they will never leave the state-cache.

You keep saying that a lot of protocols are NAT-unfriendly and deserve to
be shot (or what ever you do with your enemies in your jihad). It have
never occured to you it could be the opposite, that NAT came around and
broke behavior that those protocols depended on.

I know that I think about NAT.

Love