[RRG] Re: Critique of Sprite-MTU: PMTUD, fragmentation etc.

[Robin Whittle <rw@firstpr.com.au>]

Hi Fred,

Thanks very much for your response.  I linked to it from my critique
page.  I will think about this some more and respond on the list in
the next day or two.  My 5AM thoughts are:

A future version of the Sprite-MTU draft is likely to resolve my
first two concerns.

I still think sending a Packet To Big message while sending the
packet onwards is likely to cause trouble - and that my lightweight
approach of fragmenting larger outer packets (for reassembly in the
ETR, before decapsulation), at first, even if DF=1 is set in the
original packet, has significant advantages, at least in an ITR-ETR
setting.  I think being lightweight, transparent and non-confusing
(such as not sending back PTB messages with spuriously low MTU
values), while getting the packet to the ETR even in the face of
lower than maximum MTU limits, are really important goals.

Also, my IPTM proposal with "outer source address = sending host's
address" is capable of supporting Traceroute, with a moderate
upgrade of the Traceroute client software.


I would appreciate some comment on my idea of setting some baseline
of PMTU for IPv4 which is much higher than 576, with the following
arrangement:

1 - I assume that (however defined) the "core" of the Net supports
    MTUs well in excess of 1000, and in in excess of some carefully
    chosen value well above 1000, and so probably more than double
    576.  I wrote "1280" but perhaps the value needs to be somewhat
    below this.

2 - Which "core" routers have a lower basic MTU value than 1500?
    Probably none - I guess.  If there are a few, lets put them on
    notice they should lift their game by the time an ITR-ETR scheme
    is deployed.

3 - State in the ITR-ETR scheme that all ITR and ETR functions
    should be located such that they have at least a "1280"
    or whatever value MTU to the "core" of the Net.

Can anyone think of regions of the Net where some value pretty close
to 1500 would not be a good choice?  How many levels of tunneling,
and of what kinds, are likely to be encountered?  Are there in
practice MPLS tunnels which limit MTU below 1500?

You can't necessarily make such assumptions for IPv4 in your
Sprite-MTU system, because you intend to use it in many settings
other than from an ITR to an ETR.  However, perhaps some agreement
could be reached about the final ITR-ETR scheme requiring ITRs and
ETRs to have a higher MTU (1280 or maybe >1400?) to the "core" of
the Net.  Then Sprite-MTU could have a special IPv4 threshold much
higher than 576 bytes when operating in an ITR-ETR setting.


  Cheers

    - Robin


--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg