Re: draft-van-beijnum-multi-mtu-05.txt

[Philip Homburg <[e-mail]>]

>In general I'd prefer that we reuse existing mechanisms for path mtu 
>discovery, and not invent new ones purely for the case of two directly 
>connected neighbors.

I think making them the same is likely to be sub-optimal.

The nice thing about the current proposal is that the MTU is verified before
any application traffic is sent.

This not going to fly on the internet as a whole. There is no way that
busy DNS servers are first going to send ping traffic back to the client before
sending the DNS reply.

On the other hand, the model behind existing PMTU discovery is reactive. The
loss of a data packet triggers downward adjustment of the estimated PMTU.

In the context of ethernet jumbograms, such a reactive approach is likely to be
unattractive to operators. 

So I'd say that from a PMTU perspective directly connected neighbors are quite
different from the internet as a whole.