Re: Flow label versus Extension header - protocol itself

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 19-mei-2005, at 23:29, Margaret Wasserman wrote:

> I agree with Geoff -- SHIM6 will represent a major change to IPv6,  
> anyway, so adding an extension header is not unreasonable.

However, the sender putting in a header when the receiver doesn't  
need it would be unreasonable.

> My concern (once upon a time) with extension headers is that the  
> use of extension headers on some packets and not others effectively  
> presents a variable MTU to upper layers.  We need to make sure that  
> this is properly handled to avoid bad interactions with PMTU  
> discovery code.

This has nothing to do with PMTUD as MTU is the size of the packet  
including all headers. What you mean is the MSS, the maximum segment  
size. In all implementations I've seen TCP is smart enough to adapt  
its MSS to the MTU and additional headers statically. So presumably,  
it could do the same thing dymanically by just invoking the same  
mechanism for every packet. So worst case, all that's needed are  
performance optimizations.