MLS and CEF on Catalyst 6509 with SUP2/PFC2 and MSFC2.

[andrea.meconi]

IN CERTAIN CASES, TRAFFIC SENT TO THE ACTIVE HSRP ADDRESS IS SWITCHED BY THE MLS CEF INSTEAD OF BEING MSFC2 "SWITCHED". AN EXAMPLE IS A CLUSTER SERVER WITH A VIRTUAL IP FOR A SERVICE. WHEN THE MAIN MEMBER (OF THE CLUSTER) CRASHES AND THE VIRTUAL IP BECOMES ACTIVE ON THE SECONDARY MEMBER, THE SWITCH SENDS THE NEW PACKETS TO THE OLD MEMBER. THE "SHOW MLS CEF ENTRY" COMMAND SHOWS A INCORRECT IP-MAC ADDRESS ENTRY IN THE CACHE AND THE AGINGTIME (FAST OR NOT) DOES NOT WORK CORRECTLY.

CAN SOMEONE HELP ME?

THANKS.

[t.baranski]

Does the secondary cluster member send a gratuitous ARP when it takes over the virtual IP? This is typically what HA protocols do to avoid these types of stale ARP problems. If it is indeed doing so, I'd search for ARP-related bugs in whatever software version you're using and/or call TAC.

[Prashanth Krishnappa]

There are some bugs related to this. What software version are you running

http://www.cisco.com/cgi-bin/bugtool/onebug.pl?bugid=CSCds03351

[andrea.meconi]

We are running IOS (tm) MSFC2 Software (C6MSFC2-JK2SV-M), Version 12.1(12c)E5, EARLY DEPLOYMENT RELEASE SOFTWARE (fc1), and I don't known if there is a gratuitous arp. Related at the bug I'll try to set the cam agingtime to 10 seconds, because the arp cache is good.

Thanks.