On Mon, May 18, 2009 at 1:51 PM, Joe Touch <[email protected]> wrote: > I.e., reporting the "large majority" MTU ought to be appropriate, rather > than claming down on the smallest MTU on the path.
I think this would only be appropriate if the endpoint fragments, not drops, packets that are too big to transmit, but smaller than the reported MTU. Then choosing the MTU becomes largely a cost optimization problem. Dropping some (at best) random fraction of packets seems like asking for trouble. It would also break rfc4821. -John
