-----Original Message----- From: Robert Crowe [mailto:rocr...@cisco.com] Sent: Thursday, January 13, 2005 2:14 PM To: Matt Bazan; 'Church, Chuck'; cisc...@puck.nether.net Cc: rocr...@cisco.com Subject: RE: [c-nsp] cisco 3750 arp timeout

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

Matt,

I believe I ran into the same issue your talking about awhile back when I was doing systems work. Our issue was with pairs of servers running Win2k Advanced Server and Microsoft clustering. The storage was EMC. It ended up being the way Microsoft clustering sends the gratuitous arp. One way to tell is to failover the services and immediately do "clear arp" on the switches they are plugged into. We ended writing a script that would snmp poll the virtual address for the hostname of the box. When a failover occurred the hostname would change and the script would telnet to the 2 switches and clear the arp entries.

- - Robert Crowe

- -----Original Message----- From: cisc...@puck.nether.net [mailto:cisc...@puck.nether.net] On Behalf Of Matt Bazan Sent: Thursday, January 13, 2005 4:49 PM To: Church, Chuck; cisc...@puck.nether.net Subject: RE: [c-nsp] cisco 3750 arp timeout

Hi Chuck, In this particular case I've got two 3750's in a stack and several MS clusters attached (one cluster node plugs into stack member 1, the other node into member 2) and I'm having communication problems reaching the cluster virtual servers when I 'roll' the resources over from one node in the cluster to the other. One of my theories is that the stack could be having problems with the grat arps it's getting from the MS cluster.

As these systems are in our data center I'll have to wait till the next time I'm over there to put a sniffer on the wire to see if I can glean more info.

In the meantime, to attempt to rule out a grat arp issue, I was trying to get the ciscos to not cache the cluster's virt server IP and associated MAC.

Matt