Re: disagreement on which OS should change

[Gyan Mishra <[e-mail]>]

Fernando and Brian

I agree with what both of you are saying.


Post inline


On Sat, Apr 27, 2019, 5:02 AM Fernando Gont <[e-mail]> wrote:

> On 27/4/19 09:54, Brian Carpenter wrote:
> > I don't understand this discussion.
>
> Sigh.. I thought it was just me & insufficient amounts of caffeine :-)
>
>
> > LL addresses exist in the absence of
> > any routers and are created spontaneously by hosts with no external
> > inputs. Therefore, there is no such thing as an identity for a link,
> > there is only an interface identifier which is strictly local to the
> > host.
>
> The IID ends up being meaningful link-wise -- that's why there's (at
> least in theory) DAD for LLs.
>
> Gyan - I agree the link local can exist in the absence of any routers so
> that's a huge issue with subnet id.  So I see the problem now with subnet
> id.  This now gets very complicated as the router had to have a special
> auto config SLAAC with a DAD algorithm to allocate the subnet prefix to
> hosts similar to global unicast SLAAC auto config.  So in the absence of
> any routers the hosts would not have any link  identity until a router came
> up and then the this new link local SLAAC with DAD would initiate and
> assign the hosts on the subnet a auto configured subnet id prefix giving
> all the hosts  now link significance and now being on the same link local
> subnet dolled out by the router via this new link local SLAAC process.
>
> > Even in a point to point case, there is no reason that the two
> > hosts would agree about the link's identity.
>
> In fact, if one assumes /64, and selects non-zero bits for the rest of
> the prefix, then either you have to come up with a subnet-id
> election/disovery mechanism/algorithm, or you just can use the LLs,
> since most other nodes would probably be considered off-link.
>
> An alternative is to introduce an onlink route for fe80::/10... at which
> point the rest of the bits become, in practice, part of the IID.
>
>
> Gyan -  this is a simpler approach alternative option  if we made the 54
> bits part to the station id 118 bits as one large station  id field which
> is already unique by the EUI64 station id and we can still use the
> existing  link local DAD process.
>

Cheers,
> --
> Fernando Gont
> SI6 Networks
> e-mail: [e-mail]
> PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492
>
>
>
>
>