We have encountered a very strange problem with
Cat 6506:
SupII PFC2/MSFC2 2GE
WS-6148-RJ45
WS-6408A-GBIC
we asked for Native IOS on the box and got 12.1(19)E1
which we upgraded to 12.1(19)E1a. Bench testing looked OK
but when we cut it into service routing fell to pieces.
The minimal criminal is two Layer 3 VLAN interfaces:
one having a port on the WS-6148-RJ45 and a single host attached,
the other including a GE port on the Supervisor (1000BaseT) connected
to a Layer 2 network (3550-12T)
Pinging from the host on the WS-6148-RJ45 to a host on the subnet
connected to the GE port fails if the destination host is not already
in the router ARP cache. debug arp did not show the router issuing the
ARP request. If I sit on the router and issue the ping the ARP cache is correctly
issued and subsequent connectivity is OK, until the ARP is flushed.
Bizzarely if the GE link is broken out on to a 10Mbps hub
(we did this to look for the ARP request on a 10Mps monitor) the router
will ARP for the destination. Breaking the GE link down to 100Mbps the
router does not ARP. After the breakdown to 10Mbps, breaking back to
GE the router does continues to ARP as would be expected.
Doing a shut/ no shut on the Layer 3 VLAN restores the broken behaviour.
Jumping branches to 12.1(13)E11 fixes the problem and we have gone
with that in production so I cannot do any further repros of the problem
but I thought I would ask on the forum if anyone else had seen anything
like this. The only difference I spotted between 12.1(19)E1a and 12.1(13)E11
that might be significant is that the former showed dCEF in the CEF flags
under 'sh ip interface'
Lee Flight
Network Support, Computer Centre
University of Leicester