Howdy everyone,

I am having an arp problem on one of my switches, a cat 6509. 
(Technically I suppose it would be with the MSFC...)

Here's my setup:

M$ Cluster(Vlan1) --> 6509ios <-trunk-> 6509cat --> Users(Vlan2)

The switches are flat, both have all Vlans. 6509ios is the gw for Vlan1 
and 6509cat is the gw for Vlan2.

We have an M$ cluster set up and being tested, we found that during a 
failover the systems on the same VLAN seem to be okay with picking up 
the new MAC address fairly quickly, they miss one ping, but crossing 
VLAN's does not work until I clear the ARP cache in the switch.

It would appear as if the cluster sends a grat arp and the 6509ios 
recieves it but the 6509cat never see's it. I'll try to hit it with a 
sniffer later to be sure.

  --
Laugh at your problems; everybody else does.

Ah ha, found the exact bug reported:
http://support.microsoft.com/?id=244331

Now, how to fix....

  --
Laugh at your problems; everybody else does.



Wil Schultz wrote:
> Howdy everyone,
> 
> I am having an arp problem on one of my switches, a cat 6509. 
> (Technically I suppose it would be with the MSFC...)
> 
> Here's my setup:
> 
> M$ Cluster(Vlan1) --> 6509ios <-trunk-> 6509cat --> Users(Vlan2)
> 
> The switches are flat, both have all Vlans. 6509ios is the gw for Vlan1 
> and 6509cat is the gw for Vlan2.
> 
> We have an M$ cluster set up and being tested, we found that during a 
> failover the systems on the same VLAN seem to be okay with picking up 
> the new MAC address fairly quickly, they miss one ping, but crossing 
> VLAN's does not work until I clear the ARP cache in the switch.
> 
> It would appear as if the cluster sends a grat arp and the 6509ios 
> recieves it but the 6509cat never see's it. I'll try to hit it with a 
> sniffer later to be sure.
> 
>  --
> Laugh at your problems; everybody else does.

Ah ha, found the exact bug reported:
http://support.microsoft.com/?id=244331

Now, how to fix....


Well, you could reduce the arp timeout  value on the interface that has
the MS cluster. This may not be advisable if it is a large segment, but
it will reduce the time that the router takes to age out arp entries.

In interface configuration mode:
arp timeout <seconds>

Don't use zero for the amount of seconds, zero tells the router to
never age arp entries. Set it to whatever you are comfortable with for
failover response time from your cluster. Remember it will be a balance
between performance and failover time. Default is 4 hours. So changing
to 10 seconds is a big deal. And again, probably not something I would
do if this is a large segment.

Kevin