[K12OSN] Re: nfs troubles

["Terrell Prudé Jr."]

Jon Harder wrote:
> I don't think my problem has to do with spanning tree. The ProCurve 1800
> switch does not use it.
>
> The switch does have a configuration option called "flow control" that
> helps considerably. Once I turn that on, all of my old iMacs will get
> past the NFS mount part and come up to the welcome screen.
>
> Once they are up and running, I can log users in, but any intense application
> like firefox or openoffice will cause it to freeze or reboot (except for
> a newer slot-loading model which seems to work just fine).
>
> Suggestions elsewhere indicate that this problem might be solved by reducing
> network traffic between the server and client by ensuring the the clients
> connect to NFS using TCP instead of UDP, but I don't know how to do that.
> Is that avenue worth a try?

I'd still say there's something wrong at the lower OSI layers.  Too many
thousands of people, including me, use NFS across UDP and have for many
years.  UDP-based NFS *works*.  Remember that the reason for TCP is to
enable retransmissions at Layer 4 if needed, transparently to any
applications (including NFS).  My edumacated guess would be that
packets/frames might be getting dropped somewhere, in which case using
TCP for NFS is merely a Band-Aid (hence your freezes/reboots).  There's
no real way to tell without sniffing one of the ports.

Also remember that, as far as the thin client is concerned, an "intense"
application is not one that uses lots of resources on the *server*. 
Rather, it's one that does lots of screen refreshes on the *client*. 
Running TuxType or ChildsPlay would be a good test of this (lots of X11
metadata xfer).  One way to eliminate Firefox, OpenOffice.org, or any
other big app as a problem child would be to run these apps from the
server's own keyboard/mouse/monitor and see if you can repro the behaviour.

I'd also try swapping out another switch of the same model.  If that
doesn't fix it, then I'd try another vendor's switch.  IIRC, you've
already tried a little cheapie NetGear, which made the problem go away. 
Is your ProCurve 1800 a managed switch?

--TP