Re: [v6ops] new draft: draft-v6ops-jaeggli-pmtud-ecmp-problemv6op6

[joel jaeggli <[e-mail]>]

On 2/17/14, 11:57 PM, Ray Hunter wrote:
>> From: <fred at cisco.com>
>> To: v6ops at ietf.org
>> Cc: draft-v6ops-jaeggli-pmtud-ecmp-problem at tools.ietf.org
>> Date: Mon, 17 Feb 2014 05:45:16 -0800 (PST)
>> A new draft has been posted, at
>> http://tools.ietf.org/html/draft-v6ops-jaeggli-pmtud-ecmp-problem.
>> Please take a look at it and comment.
> 
> I have read this draft.
> 
> Isn't the obvious brute force mitigation to clamp all of your DC to IPv6
> minimum MTU?

That is an obvious brute force mitigation.

It does however doom ipv6 to the minimum mtu until I decide pmtu is no
longer necessary, which may or may not be never.

There wouldn't be much point if this happened frequently but as it is,
it happens at a rate of a handful of packets per second out of a few
gig's of v6 traffic. That says a lot I think about deployment.

> And isn't it yet another hint that PMTUD really needs to be incorporated
> into the transport layer [if we needed another]?
> 
> e.g. by referencing Packetization Layer Path MTU Discovery (PLPMTUD)
> http://tools.ietf.org/html/rfc4821

Sure I agree that in-band signaling would be way less likely to interact
badly with ecmp forwarding decisions.

> That should also be handled correctly by the load balancers, or not?
>