Re: [v6ops] PMTUD issue discussion

["Templin, Fred L" <[e-mail]>]

Hi Brian,

> -----Original Message-----
> From: Brian E Carpenter [mailto:[e-mail]]
> Sent: Friday, September 05, 2014 6:41 PM
> To: Templin, Fred L
> Cc: Joe Touch; Fred Baker (fred); Fernando Gont; [e-mail]; Tom Perrine; Fernando Gont
> Subject: Re: [v6ops] PMTUD issue discussion
> 
> On 06/09/2014 02:48, Templin, Fred L wrote:
> >
> >> -----Original Message-----
> >> From: v6ops [mailto:[e-mail]] On Behalf Of Joe Touch
> >> Sent: Friday, September 05, 2014 7:00 AM
> >> To: Fred Baker (fred); Fernando Gont
> >> Cc: [e-mail]; Tom Perrine; Fernando Gont
> >> Subject: Re: [v6ops] PMTUD issue discussion
> >>
> >>
> >>
> >> On 9/4/2014 5:58 PM, Fred Baker (fred) wrote:
> >>> On Sep 5, 2014, at 10:47 AM, Fernando Gont <[e-mail]> wrote:
> >>>
> >>>> On 09/04/2014 09:24 PM, Fred Baker (fred) wrote:
> >>>>> On Sep 5, 2014, at 9:33 AM, Fernando Gont <[e-mail]>
> >>>>> wrote:
> >>>>>
> >>>>>> What happens when/if such lins employ v6 (whether they really
> >>>>>> offer a "virtual" MTU >=1280).. I just don't know.
> >>>>> 802.15.4?
> >>>> No. I seem to recall that it was more of a packet radio sort of thing
> >>>> (something predating 802.15.4). I could ask some folks if interested
> >>>> in more details...
> >>> I think you missed my point. We have a case in point today, which is
> >>> IPv6 in an IPv6 tunnel encapsulation running on a packet network whose
> >>> MTU is 127 bytes. We call it "6lowpan", and it runs on IEEE 802.15.4.
> >> That's not IPv6. I don't know what it is, but RFC2460 is clear on the
> >> minimum MTU for IPv6.
> >
> > Joe is right - for IPv6 there is only one magic number (1280). Beyond
> > that, any size up to and including jumbograms must be accommodated,
> > subject to path MTU limitations.
> 
> Actually jumbograms, according to RFC 2675, are only available
> when the *link* MTU is big enough:
> 
> "  The Jumbo Payload option is relevant only for IPv6 nodes that may be
>    attached to links with a link MTU greater than 65,575 octets (that
>    is, 65,535 + 40, where 40 octets is the size of the IPv6 header).
>    The Jumbo Payload option need not be implemented or understood by
>    IPv6 nodes that do not support attachment to links with MTU greater
>    than 65,575."
> 
> Note that it says "nodes", so it applies to routers as well as hosts.
> My recollection is that efficent use of HIPPI within data centres
> was the main motivation for jumbograms, and there is no expectation
> that they will work on the Internet in general. Running jumbograms
> over an adaptation layer would be craziness.
> 
> Also
> "  The Jumbo Payload option must not be used in a packet that carries a
>    Fragment header."
> 
> So I think it's "any size up to 65,575 octets" that must be
> accomodated. Jumbograms are optional.

The key phrase in my text that you quoted was "subject to path MTU
limitations". For example, if the source's IPv6 layer presents a 4MB
packet, and if all links in the path support an MTU that is at least
that large, then the packet will be forwarded to the destination just
fine. If the path MTU is deficient at any hop along the path, however,
the packet is dropped and a PTB message returned.

We may be a long way off from having paths in the Internet (or even
in private networks) that supports such large sizes, but the protocols
will be ready when and if such links come online.

Thanks - Fred
[e-mail] 

>   Brian
> 
> 
> >
> > Thanks - Fred
> > [e-mail]
> >
> >> Anyone who pays money for such a thing should return it.
> >>
> >> Joe
> >>
> >> _______________________________________________
> >> v6ops mailing list
> >> [e-mail]
> >> https://www.ietf.org/mailman/listinfo/v6ops
> >
> > _______________________________________________
> > v6ops mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/v6ops
> >