enabling IPv6 on Cisco 6500 breaks IPv4 Internet connectivity

Phil Mayers p.mayers at imperial.ac.uk
Sun Jun 10 12:58:58 CEST 2012

On 06/10/2012 12:31 AM, Jim Trotz wrote:
> Here is an update:

A lot of information there, but I'm afraid it's not very... systematic.

If traffic is flowing but going slow, as other people and myself have 
pointed out, the most likely cause is it's being punted to CPU.

You say the CPU usage isn't rising: this might be because the CPU punted 
traffic is being rate-limited before reaching the CPU by CoPP.

Are you running CoPP? Does your CoPP policy by any chance limit UDP/TCP 
but not ICMP? Of course, there's likely less ICMP, so even if it's 
limited but separately, this could be the cause.

> The commands are 'ipv6 unicast-routing' and adding 'ipv6
> address-family'. Added separately - no problem. Both together - problem.

You keep saying "ipv6 address-family". I don't think there's any such 
command. Do you mean:

ipv6 unicast-routing
vrf definition XXX
   address-family ipv6


> Further lab testing shows that IPV4 ICMP & UDP & TCP traffic still pass
> through the router during the "hang" period. IPV4 UDP & TCP traffic
> comes to a crawl (0-2000bs/sec).

But not ICMP.

If UDP/TCP are going slow but ICMP isn't, then I suspect you are seeing 
CPU punts, modulated by CoPP.

> Also, noticed that if I shutdown the uplinks to the Internet Hubs then
> the problem goes away, even before BGP neighbors go into ACTIVE status.
> (The data path in my tests does not use these uplinks, just ports on
> trhe test box)

What is an "internet hub"?

Have you tried using SPAN to see what's hitting the CPU during the 
hang/outage/whatever we're calling it?

Can you confirm you don't have IPv6 uRPF enabled anywhere?

I think you're punting to CPU. I think you need to find out why.

It *could* be a TCAM mis-programming bug in IOS. It could be a feature 
you've got enabled - uRPF, ICMPv6 redirect & a routing loop, etc.

More information about the ipv6-ops mailing list