SUMMARY: Sol8 / netraX1 / packet loss

The cause of the problems has been tracked down to a failure of
network autonegotiation between the netras and the cisco switch.
The netras decided they had 100MHz full duplex, the switch decided 10MHz
half duplex.

Why autonegotiation worked OK for months and still works for 7 out of 10
identical netra clients is still a mystery.

The solution would appear to force 100MHz full duplex on the netras and the
switch.  Now all I need do is track down the person who initially set up the
switch and so kindly set a password which I now need to force the settings :-)

Thanks for the replies 
	Jose Marcos Lopez Caravaca
	Geoff Reed 
	Todd M. Wilkinson
	and all those who were fortunate to be already away for christmas.



On Fri, Dec 19, 2003 at 11:53:08AM +0000, Geoff Lane wrote:
> I've got a network problem that seems intractable.
> 
> We have a NFS server (E4500/Solaris 8) connected to ten clients (Netra
> X1/Solaris 8) via a Cisco Catalyst 2950 switch.  We use Fibre between the
> server and the switch (via the GBIC interface) and 100MHz Cat5 between the
> switch and the NetraX1s. IPV6 is enabled on all machines, no machine acts as
> a router. The switch forms a private network used only for NFS traffic.
> All the clients were installed using jumpstart and a common OS image.  
> 
> This has been working well for about a year.
> 
> A few weeks ago we noticed NFS problems on two of the clients.  Mounting the
> NFS volumes was slow, as was accessing the data on the mounted volumes.  On
> investigation (with the aid of Sun), nothing could be found wrong with the
> client network hardware or software.  The problem clients also have
> difficulty using the private network to access other clients; we see 90%
> packet loss when pinging between a problem client and a working client over
> the private network even when it is otherwise idle.
> 
> We've power cycled the switch, rebooted the clients and restarted the NFS
> daemons on the server with no improvement.  We've replaced the Cat5 cables
> to the problem clients. We've moved the clients to different switch ports.
> The second NIC on each Netra is working without any problems.
> 
> Running snoop on the problem client and another system indicates that the
> packets do reach the switch.  The symptoms suggest that the switch is
> dropping about 90% of the packets originating from just two out of 10
> clients.
> 
> Can anybody suggest anything else to try on Solaris before we try replacing
> the switch?
> 
> Thanks,
> 
> -- 
> /\ Geoff. Lane. /\ Manchester Computing /\ Manchester /\ M13 9PL /\ England /\
> 
> "Bother", said Pooh, as the vice squad took his GIFS
> _______________________________________________
> sunmanagers mailing list
> sunmanagers@sunmanagers.org
> http://www.sunmanagers.org/mailman/listinfo/sunmanagers

-- 
/\ Geoff. Lane. /\ Manchester Computing /\ Manchester /\ M13 9PL /\ England /\

"Bother", said Pooh, as he failed the dope test.
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers