Re: I-D Action: draft-voyer-6man-extension-header-insertion-07.txt

[Tom Herbert <[e-mail]>]

On Mon, Oct 21, 2019, 1:34 AM Alexandre Petrescu <
[e-mail]> wrote:

>
>
> Le 18/10/2019 à 16:56, Tom Herbert a écrit :
> > On Fri, Oct 18, 2019 at 2:23 AM Alexandre Petrescu
> > <[e-mail]> wrote:
> >>
> >>
> >>
> >> Le 17/10/2019 à 18:19, Tom Herbert a écrit :
> >>>
> >>>
> >>> On Thu, Oct 17, 2019, 6:20 AM Alexandre Petrescu
> >>> <[e-mail] <mailto:[e-mail]>>
> wrote:
> >>>
> >>>
> >>>
> >>>      Le 15/10/2019 à 00:12, Fernando Gont a écrit :
> >>>       > On 12/10/19 16:57, Brian E Carpenter wrote:
> >>>       >> Hi,
> >>>       >>
> >>>       >> I'd like to comment on this version. It is in fact a complete
> >>>       >> rewrite compared to its predecessors and I thank the authors
> for
> >>>       >> that. The tone is now purely technical, and that's a great
> >>>       >> improvement.
> >>>       >
> >>>       > It is somewhat frustrating that the draft still fails to argue
> why
> >>>       > EH insertion instead of encapsulation.
> >>>
> >>>      As usual, I think one of the reasons is a difficulty in
> qualifying what
> >>>      it means 'encapsulation'.
> >>>
> >>>      There is IP-in-IP encapsulation.
> >>>
> >>>      But there is also encapsulation like in transporting, or
> carrying, by
> >>>      means of other intermediary headers, layer2, MPLS, security
> headers and
> >>>      future internet shims and GRE and routing headers.
> >>>
> >>>      IP-in-IP encapsulation is clearly an alternative to EH insertion.
> >>>
> >>>      But all the other encapsulations are so messy that one may
> legitimately
> >>>      think that a new EH insert/delete standardized according to good
> WG
> >>>      principles would be proper, universal, and solve all  problems of
> GRE
> >>>      for example.
> >>>
> >>>
> >>> Alex,
> >>>
> >>> What are the problems of GRE to which you referring?
> >>
> >> Tom,
> >>
> >> If I remember correctly, on the negative side, GRE does not work with
> >> IPsec security, GRE has no IPv6-GRE-IPv6 deployment, GRE is for
> >> limited-domains and does not work across the Internet contrary to VPN.
> >>
> > Alex,
> >
> > I guess maybe you're only considering GRE/Ethernet, but for GRE/IP I
> > don't believe any of these negatives are valid. In fact, an IPv6
> > packet that contains next header 47, a 4 byte GRE header with first 16
> > bits set to zero and Protocol type set to 0x86dd, followed by an IPv6
> > packet-- is functionally equivalent to ip6ip6 encapsulation. Also, if
> > you want to increase the probability of successful traversal through
> > the Internet then GRE/UDP can be used (RFC8086).
>
> One may wonder whether IP/GRE/Ethernet, or GRE/UDP/IP, represents
> encapsulation that avoids the necessity of EH insertion, or does not
> represent such.
>
> >> On the positive side, thanks to its key field, only with GRE it is
> >> possible to link together several tunnels such as to increase available
> >> bandwidth by aggregating, whereas with IP-in-IP it is not possible.
> >>
> > Yes an advantage of GRE over just ipip encapsulation is that it's
> > extensible, albeit limited in its extensibility which is why we
> > developed GUE.
>
> I meant an advantage more issued from implementation, not from
> specification.
>
> The implementation of ip-in-ip tunnelling in linux is with virtual
> interfaces.  These interfaces have only two parameters: local and remote
> address.  The GRE implementation uses a  key behind the scenes, in
> addition to these same two parameters.  That key is what helps a
> bandwidth augmenting implementation to keep up  two tunnels at the same
> time (same 'remote' addresses, but different 'local' addresses), to
> distinguish between them, and to distribute IP packets between the two
> (round-robin is one such fashion).  The ip-in-ip tunnel implementations
> do not allow such distribution, because when you put the same 'remote'
> address on two ip-in-ip interfaces it will use just one such interface.
>
> In such a context, one could imagine using EH insertion with IP-in-IP
> tunnelling: insert a key (in an EH) dynamically, by the endpoints.  This
> could help IP-in-IP classic tunnelling to act like GRE tunnelling, with
> respect to the apps that need to maintain two tunnels at the same time.
>

Alex,

You're conflating protocol specification with protocol implementation. It's
much better to change the implementation to leverage the protocols, rather
than to change the protocols to suit the implementation.

Tom


> Alex
>
> >
> > Tom
> >
> >> Alex
> >>
> >>>
> >>> Tom
> >>>
> >>>
> >>>      Alex
> >>>
> >>>       >
> >>>       >
> >>>
> >>>
> --------------------------------------------------------------------
> >>>      IETF IPv6 working group mailing list
> >>>      [e-mail] <mailto:[e-mail]>
> >>>      Administrative Requests:
> https://www.ietf.org/mailman/listinfo/ipv6
> >>>
> --------------------------------------------------------------------
> >>>
>