Robert,

  Thanks for the info.  I'll put on my queue to try the kernel code in
the 4.7.4 tree.  I know it's in 3.0, but I have to run 2.2.
It sounds like there has been no interface change between kernel in
user from 4.6.5 to 4.7.4.

  4.7.x code included many bugfixes and race conditions concerning the death

Good news; things are generally ok except when venus gets so wedged I
need to kill it.

  Coda currently does not behave well when clients are behind a firewall.

I should have been more clear.  I meant by firewall a 'filtering
router', not NAT box.  The actual IP address of the client shows up at
the server, unmunged.  Packets are either let through or dropped, but
not modified.  I see your point about NAT boxes, but fortunately
don't have any.  I have seen the callbacks, but that's not my problem
at the moment.

  package based on UDP.  The use of encryption also prohibits firewall
  involvement, as it makes it difficult for the firewall to modify packets

It prohihits NAT/proxy, but not a filtering one.  If coda encrypts
inside udp, the port numbers will be present and most filters will be
ok.  AH/ESP would be more problematic.

  There used to be some strange failure conditions when the main RPC
  connection succeeded and the side effects consistently failed -- have you

I observe venus retrying and not giving up (like I think it should).

I'm concentrating on client in, server out, no NAT, as that gets me
what I want, which is a bunch of clients, in and out, being able to
talk to the same server.  UDP priming may work, perhaps in rpc2.

        Greg Troxel <gdt_at_ir.bbn.com>