Re: [rohc] Path MTU discovery when having varying MTU

On May 26, 2009, at 10:51, Magnus Westerlund wrote:

> One end point sends ROHC segmentation, the other end
> doesn't implement it

We did not really make segmentation optional in RFC3095/RFC4995, but  
setting MRRU to 0 effectively disallows the use of segmentation, so  
its use can indeed be negotiated.  Enabling the use of ROHC  
segmentation in HCoIPsec would entail supporting negotiation of the  
MRRU option in IKE-ROHC (which is currently hardwired at 0).

Of course, options and negotiation do not necessarily aid  
interoperability.  But then, "extending the MTU" would be an  
optimization, and implementations can always fall back on the worst- 
case MTU*) approach Joe suggested, if the tunnel MTU is large enough  
to support the minimum MTU.

Gruesse, Carsten

*) I haven't done the work yet to verify that there is a good way to  
compute this worst-case MTU.
But I think it feels right.