Incorrect arp response for Fast0 interface on 3020

[alanjbrown]

We have a lot of  Cisco 3020 (WS-CBS3020-HPQ) but one pair in an HP enclosure have been providing very strange ARP entries.

The blade module has 1 port channel down stream to a 4948 and is connected to serveral blades servers upstream. Management is provided by a Fastethernet0 port which sees the other blade module and our iLO switch. We have one vlan (Vlan 100) configured and initially the fast0 port was getting it's ip address via DHCP configured from the enclosure. This problem started about a week ago and so we have now assigned the address statically and the default gateway for fast0 to see if this would resolve the issue.

The Fast0 address is 10.60.67.49/22 and default gateway is 10.60.64.1. We are now no longer able to directly connect to the 3020 and it's adding strange entries into the arp table.

In a normal 3020 we see a table as follows:

Internet  10.60.64.2             27   0024.f71d.dac1  ARPA   FastEthernet0

Internet  10.60.64.3             15   0024.f719.cdc1  ARPA   FastEthernet0

Internet  10.60.64.1             72   0000.0c9f.f096  ARPA   FastEthernet0 -- HSRP address

In the faulty 3020 we see:

Internet  10.60.64.2             18   0024.f71d.dac1  ARPA   FastEthernet0

Internet  10.60.64.3              0   0024.f719.cdc1  ARPA   FastEthernet0

Internet  10.60.64.1             10   001b.531b.2a41  ARPA   Vlan100

The Vlan 100 is in address range 10.60.1.0/24

How is is possible that the default gateway arp entry is being learned via vlan 100 which is in a completly different ip address range?

thks,

Al

[lgijssel]

This question is hard to answer without more detailed info regarding the topology.

Perhaps a native vlan mismatch?

However, the best approach is to find out where the mac address is originating from.

This could provide a first clue to this peculiar issue.

regards,

Leo

[alanjbrown]

Leo,

Thanks for replying so this is where it starts to get interesting. The address 001b.531b.2a41 is actually for address 10.60.1.22 which is on another blade switch on another HP enclosure.

Topology is easy:

Servers -- Blade Switch -- Port-channel -- Cisco 4948 -- Nexus 7000.

thks,

Alan

[lgijssel]

alanjbrown wrote:
Topology is easy:
Servers -- Blade Switch -- Port-channel -- Cisco 4948 -- Nexus 7000.

But the link between Blade Switch and C4948 is probably a trunk, i.e. using multiple vlans?

This really smells like native vlans being messed up.

Can you provide some stp related output?

For a start sh vlan, sh spanning-tree vl 100, sh spanning-tree brief ?

regards,

Leo

[alanjbrown]

Leo,

The connection is a trunk carrying multiple vlans. They have the same native vlan configured at both ends.

BladeModule1

interface Port-channel1

description to 4948 - Po3

switchport trunk encapsulation dot1q

switchport trunk native vlan 999

switchport trunk allowed vlan 106,100

switchport mode trunk

4948

interface Port-channel3

description Blade Module 1 - Po1

switchport

switchport trunk encapsulation dot1q

switchport trunk native vlan 999

switchport trunk allowed vlan 106,1000

switchport mode trunk

show vlan looks ok -- Ports which have never been used are in Vlan1 and the host servers are in 106

show spanning looks ok -- Vlan 106 desg ports to server and root upstream via Po1 and Vlan 100 only root upstream via Po1 to 4948.

My strongest feeling is a bug but until we reload; I want to do as many checks as I can or even identify the bug and so I can recommend an upgrade.