Re: [pmtud] Improvement for the current PMTUD mechanism

On Tuesday 11 October 2005 09:22 am, Fernando Gont wrote:
> At 05:00 p.m. 10/10/2005, John Heffner wrote:
> >One more comment that just occurred to me.  The ICMP checking fails in the
> >case of gear that doesn't honor the DF bit (that is, sends an ICMP but
> > also fragments and forwards the packet despite the DF bit).  This is an
> > issue we discussed at some pmtud meetings.
> >
> >I think such gear can be considered broken for a number of reasons (and is
> >certainly not standards compliant) but does supposedly exist in the wild. 
> > It might be worth documenting this case.
>
> As stated in my draft, my proposal does not address the issue of blackholes
> or broken implementations.
> It's an *improvement* for the current PMTUD mechanism, rather than a
> replacement for it.

This is not a black hole issue.  In this case, classical pmtud works fine but 
adding in a loss check causes pmtud to "fail" such that all packets get 
fragmented.  This is not a total connection failure like a black hole, but is 
bad for a number of reasons.

http://www.hpl.hp.com/techreports/Compaq-DEC/WRL-87-3.pdf
http://www.psc.edu/~jheffner/drafts/draft-mathis-frag-harmful-XX.html

  -John

_______________________________________________
pmtud mailing list
pmtud@ietf.org
https://www1.ietf.org/mailman/listinfo/pmtud