[c-nsp] Problem with irregular lost contact between 2821 and 2950

[Peter Olsson]

We have a stub LAN consisting of a 2950T-24 switch with one server.
The 2950T-24 switch is connected to a 2821 router.

Both the 2950T-24 and the 2821 were installed about a month ago, and
there have been nearly ten occasions of lost contact since then.
This happens mostly at night, and the pattern seems to be that a backup
job is started on the server and this makes the server lose contact
and a short while after that the 2950T-24 also loses contact.
They both disappear from the ARP table in the 2821, and they don't
come back until the router or the switch is rebooted, or the router
interface toward the switch is shutdown and then enabled again.

Debug shows that the switch receives ARP questions from the router,
and it also sends ARP replies to the router, but the router doesn't
seem to receive the ARP replies.

We have a cron job sending three pings each to the switch and the server
every minute, but this doesn't stop them from disappearing from the ARP
table in the router.

It seems that there is some bug in the 2821, probably triggered by
a heavy load (backup job or something similar).
I searched for 12.4(8) "arp" in Bug Tool but found nothing interesting.

We have tried both with and without ip cef in the 2821, and we have
tried with these IOS:

c2800nm-advsecurityk9-mz.124-8
c2800nm-advsecurityk9-mz.124-7a

c2950-i6q4l2-mz.121-22.EA8
c2950-i6q4l2-mz.121-22.EA7

The 2821 runs OSPF, CBAC and IPSEC, but not on the interface toward
the 2950T-24. OSPF, CBAC and IPSEC are used on VLAN:s on the other LAN
interface, which is connected to "the outside world", in the 2821.

We have auto speed and auto duplex between the router and the switch.
Both show 1000Mb/s Full-duplex.

Any ideas before we open a TAC case?

Thanks!

-- 
Peter Olsson                    pol at leissner.se