[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Fwd: Re: ISP IPv6 Deployment Scenarios in Broadband Access




Hi Gert,

Few comments in line (marked @@) ...

At 11:17 AM 11/23/2004 +0100, Gert Doering wrote:
Hi,

On Tue, Nov 23, 2004 at 11:58:13AM +0200, Pekka Savola wrote:
> >>>Alternatives seem to be:
> >>> 1) defining each /64 prefix to advertise for each customer manually,
> >>> not using bulk methods: this does not scale, so it's not an option.

This is what we do on leased line customers (or ATM point-to-point PVC DSL
access).

> >>> 4) putting all the customers' v6 prefix information in a RADIUS or
> >>>similar database, so that the advertisement information could be digged
> >>>up from there. A lot of work, and does not work automatically. This
> >>>would also need some glue between bulk config and RADIUS.

This is what we do for "bulk" DSL access, which is presented to us as
L2TP connections, so RADIUS is the "natural" way to do it.

@@
Yes, this the way to do it for a scalable deployment. One way to look at the option mentioned by Pekka:


" 2) checking whether providing a single, shared /64 for all the customers would work (what if there are address conflicts? what about communication inside the prefix? does this work?)"

is to see it as a NBMA network made of multiple PPPoE sessions terminated on the same edge router where the initiators of the sessions are assigned addresses from the same /64 i.e. similar to IPv4. If the users are authenticated locally and provided addresses from the pools defined on the router then the RADIUS is not needed and for this reason we mentioned it here and not under option 4). This option is simply for the sake of discussing the perspective without bringing in the RADIUS but for a production deployments, RADIUS is the natural way to do it as you mentioned.

Pekka, did you have something else in mind with option 2), something for which the PPP NBMA model discussed in the draft is not representative? Something more down the lines of IPv4 RBE like environment?
@@



[..]
> >You might have see Gert Doering's email on this that option 2 works well.
>
> Ummm.. As far as I saw from Gert's mail, he was doing 4) (or possibly
> something something close to 1), not 2) ?

Yes, there must have been some misunderstanding.

Option 2) is something we can't easily test, as we only have a handful
of DSL lines with bridged ethernet available, and on those, our equipment
can't do IPv6 RBE (Cisco 12.3, not 12.3T).  Doing it with a bridge-group
might work, but definitely doesn't scale.

@@
One option would be to try IRB but that definitely is not a solution for a deployment. As far as IPv6 RBE is concerned, I would like to point out that IPv6 RBE is conceptually different then IPv4 RBE. While in the case of IPv4 the PVCs belong to the same NBMA, same IPv4 subnet, in the case of IPv6 RBE each PVC has its own prefix and the traffic is routed between the PVCs.
@@



Gert Doering
        -- NetMaster
--
Total number of prefixes smaller than registry allocations:  66629  (65398)

SpaceNet AG                 Mail: netmaster@Space.Net
Joseph-Dollinger-Bogen 14   Tel : +49-89-32356-0
80807 Muenchen              Fax : +49-89-32356-299