[vortex-bug] [DESPERATE] Need some insight on Tornado's connected to a foundry switch

[Andrew Morton]

"Ian C.Sison" wrote:
> 
> Hello anyone who can help!
> 
> I've got several servers connected to a foundry serveriron L4 switch.  They are
> dual P3s, with 3 adaptec SCSI controllers and 2 3com Tornadoes.  I only use one
> tornado, however.  The other nic is installed but it's interface is private and
> not being used.  OS is Redhat 6.2, running the stock linus (not redhat) kernel
> 2.2.16 with the combo errata patch.  The servers are a farm basically used for
> squid proxy/cache service, and have fairly heavy ethernet traffic.
> 
> Problem:  Every once in a while, the ethernet interface fails.  The server
> itself is ok (no kernel or driver errors in the logfiles), the only difference
> is that it cannot see the network anymore, both ways.  Ping fails, and believe
> it or not, i cannot even ping the IP address assigned to the ethernet cards
> _from the local machine itself_ ! ! !
> 
> The only solution to this would either be to (1) reboot, or (2) restart the
> network (via /etc/rc.d/init.d/network restart) or (3) manually do an ifconfig
> eth0 down; ifconfig eth0 up procedure.  Obviously the downtime of this is
> unacceptable.
> 
> Question is why is this happening?  I am tempted to put the blame on the
> foundry switch, but i have nothing to back-up my claims.
> 
> Anyone here knows what is going on?

This is creepy.

Even if the interface has magically downed itself or if the driver has gone
to lunch, you should still be able to ping the interface address.

Next time it happens, can you run `ifconfig -a' and `route' and send us the
output?  Run `ifconfig' a few times while someone is sending packets at you
to see if the driver is actualy receiving packets.

Does it happen to just one machine, or to all of them?

Even if the routing has gone silly, `ifconfig up' shouldn't help.  Are you
sure you're not using `ifup'?