On Tue, Apr 03, 2001 at 09:16:08AM -0400, John Hawkinson wrote:
> For NFS in environments where it crosses a router, or [help us!]
> a WAN, a 32k fragmented UDP packet can be far more severe than
> an 8k one, because it may be substantially more likely
> that the drop of a single fragment will occur, for two reasons:
[...]

Yes, this is certainly true. However, such a setup is likely
a minority of situations, so not something to adapt the
default setting for. And in such cases it would be wise to
use TCP anyway.

> Have you considered these issues? I think they're important,
> perhaps important enough to get discussion in a man page. I
> note we don't have an nfs(4) -- too bad. At lesat some discussion
> under mount_nfs(8) under -r and -w, though.

I agree that some mention in the manpages would be wise. I'll try
to come up with something.

There is also one other issue (for v3): the sizes that the server
advertises. These have been traditionally too large in our code
(our client side stuck to 8k, but servers advertised 16k). Since
the client will use the minimum of the two, you will only notice
this is you have a client that can handle large packets well,
but a server that doesn't. In which case you shouldn't be using
that machine as a server, but..

- Frank

-- 
Frank van der Linden                           fvdl@wasabisystems.com
======================================================================
Quality NetBSD CDs, Support & Service.   http://www.wasabisystems.com/