On Sun, 24 Jan 1999, SOM Bandyopadhyay wrote:

> I have this question for CODA ...
> 
> 1. Is there a callback mechanism from CODA server to client?

Yes, just like in AFS, it's like oplocks in SMB protocol, but the cached
data is persistent.

> 
> 2. If so, does it happen when clients are not connected to servers? In
> that case, do CODA servers initiate TCP connections to CODA clients?

First of all, we use UDP not TCP.  Our RPC2 package has a concept of
connections, that are supposedly more light weight than TCP.  All these
connections are multiplexed over a single UDP socket.  For file transfers
we use a "side effect" called SFTP, which uses one other socket.

The servers and clients drop all callback and connection information when
the connection disappears.  When the client reconnects, it needs to
revalidate it's cache, for which it uses a fast, multilevel cache
validation scheme.

> 
> 3. If 2 is true...then I have list of problems and concerns including 

Oh dear...

> 
>             a) how do u handle DHCP address change?

Probably this works just fine.  I don't think that the IP of the client is
every stored across disconnection, as you see above.  When that client
comes back, with a different IP, the server just sees a new client, but
that is fine.  Admittedly, this may not quite work correctly at the
moment, but the what I am trying to say is that I don't see a design
problem.

>             b) firewalls limiting incoming connections..

This is a sore point.  We need to modify the UDP connection setup so that
the client also sets up the "reverse" connection for the callbacks (where
the server is on the client and the client is on the Coda server.).

>             c) limitations of Win95/98 ( may be NT workstations) to
> handle large number of incoming connections...

This is no problem, since we use a single UDP socket on the server (well,
in fact we use two) for all clients.


- Peter -

> 
> etc....
> 
> 
> thanks,   som.
>  
>