[c-nsp] how "fast" is L2TP "fast switching" under 12.3T?

[Krzysztof Adamski]

On Thu, 28 Oct 2004, Alex Rubenstein wrote:

>
>
> > Here is something you can do to eliminate the UDP problem, set the MTU on
> > the PPPoE link to 1452, now you are going to guaranty that no fragmentation
> > is not happening. I have not been brave enough to try this, since I don't
> > know what will stop working :-) If you try this, let me know how it works.
>
>
> This is not a guarantee at all. Assume for a moment:
>
> (windows host)---ethernet----(soho-router)---pppoeoatm----(ISP router)
>
>
> While the ISP router during the ppp negotiation phase will tell the
> soho-router a MTU/MRU of whatever you ask (1400, 1452 or whatever), the
> windows host has no idea what going on regarding the MTU there.
>
> You are making the assumption that PMTUD works, and, well, thats a wild
> assumption.

You are missing the point, setting the MTU on the PPPoE link to 1452 does
not eliminate fragmentation end to end, but it eliminates reassembly on
the LNS. The reassembly is the most CPU intensive task. It will still have
to fragment the returning packets going to the windows-host, but that is
less CPU intensive. Also, considering that most window boxes set the DF
bit (don't ask me why) it should stop the fragmentation.

K