Cisco Support Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Announcements

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. If you'd prefer to explore, try our test area to get started. And see here for current known issues.

New Member

why would it be necessary to clear ARP on a 6500

when 'sh ip arp' shows the correct MAC address for the 'now active nic'. Until we did a 'clear arp' all packets for an IP were being sent to the nic (on a different machine) that previously had the IP. It's been suggested that 'mls rp ip' on the vlan interfaces is the problem.

Here's the details of the problem if needed:

We have a small network with 2 7000 border routers and a single 6509 doing vlans and ospf, nothing complicated. We have several 2-machine linux clusters doing failover using a single IP aliased to eth0:0 on whichever machine is in control, nothing complicated here either. Recently, it was necessary to add a /29 to one cluster. A static route was added on the 6509 to point this /29 to the failover IP of that cluster. This weekend we had a failure of the master in this cluster and the slave came up and aquired the IP correctly and took over all services fine. We replaced a bad drive on the master and brought it back up, it reaquired the IP and all appeared to be fine except packets for the /29 were still being sent to the slave.

Thanks for any advice.

Ron

3 REPLIES
Bronze

Re: why would it be necessary to clear ARP on a 6500

It is indeed strange to see the switch forwarding it to the old MAC address when the ARP table has the new MAC address. Just wondering if changing the ARP cache timer will make any change for this behavior. You can use a network analyzer to see which address is being used to forward the frames.But, you should have faced the same problem even when the slave tookover from the master.

Bronze

Re: why would it be necessary to clear ARP on a 6500

I think the MLS is the problem. The arp table is not being used because the data is being layer 3 switched using the MLS entries in the switch, it never reaches the router(msfc). I would think that changes in ARP entries would trigger a change in the MLS table but it sounds like it did not. There might be a timer issue (aging) in the switch or a IOS issue. I know I had to clear the MLS table once by going the the MSFC and clearing the ip route table. Some of this is old because I'm on a Super1A.

New Member

Re: why would it be necessary to clear ARP on a 6500

I did some tests for a client using Novell Cluster. it appears that the MSFC doesn't

handle Gratuitious ARP quickly. I discovered (with a MAC: thanks to alternate platforms and IP stacks), that the MSFC did accept the new ARP entry after 8 gratuitious ARP.

Nonetheless, a simple ARP query did change the ARP table. But if a machine did manage several address thru a single interface, the router MAC address is always known and then, no ARP query is sent.

The workaround we found was, for the server to make a clear ARP on the server. by doing so, the server did send a ARP request and the MSFC did change its table.

I tought, at the very beginning that it was connected to MLS but: 1) it is not possible to use a timer shorter that 30 s 2) we change from one 6500/MPSFC from another (and then change the MLS tables) without noticing any change. Even a clearing of the MLS table didn't change naything.

181
Views
0
Helpful
3
Replies