Re: [v6ops] IPv6 fragmentation experience

[Joe Touch <[e-mail]>]

> On Mar 23, 2018, at 3:20 AM, Ole Troan <[e-mail]> wrote:
> 
> Joe,
> 
>>> It seems you are in denial about the consequences for IPv4 with ever increasing address sharing.
>> 
>> You are willing to pay money to be lied to. I am not
> 
> I am the one taking your money and implementing it, and trying not to lie to you in the process.

Here’s what you need to do to stop lying that you’re selling me an IP router (or A+P device):

1) if you look at L4, then you need to either reassemble the packet or cache the front fragment and use it as context for later fragment forwarding.

2) if you can’t keep up with #1, then you need to ignore L4 context for that packet and forward it using only L3

3) remember that ECMP still requires a valid P (path); if a “path” doesn’t go to a destination for a fragment, it’s not a path and should not be in your forwarding table

I’m doing what I can to help reduce the amount of #2 traffic here in the IETF (UDP options) and certainly everyone should be doing some sort of PLPMTUD too. Sure, we all should also try to move to IPv6.

But that doesn’t mean those who sell IPv4 devices get to lie.

We all need to be aware of implementation *errors* and fix them.

Joe