BCP for multisite multihoming

[Carlos Garcia Braschi]

I don't know, I still think the idea could work... In IPv6 we could
afford to do that only for multihoming customers... and it would be
somehow better than current multihoming practice.

But yes, the financial incentive is small, although multihoming
customers are usually some of the better customers. And customers get
to keep their addresses as long as they keep being "near" the IX.

2007/7/21, Roger Jorgensen <rogerj at jorgensen.no>:
> On Fri, 20 Jul 2007, Carlos Garcia Braschi wrote:
> > 2007/5/23, Brian E Carpenter <brian.e.carpenter at gmail.com>:
> <snip>
> >>
> >> BTW the same is true of geographic addressing - as Iljitsch
> >> says, it isn't hard technically. What has been lacking for the
> >> last 15 years is a financial incentive.
> >
> > I'm not sure if this has been proposed already, but what if we did
> > geographic addressing but assigned the addresses to
> > interconnect/peering points? (requiring that they be used to cover the
> > region the IX is in, and that among all the peering partners manage to
> > exchange locally the un-aggregated routing tables).
> >
> > Those would act as LIR / RIR for all ISPs connected to them and would
> > have incentive to promote the idea... as it promotes fidelity of their
> > peering customers and gives them more service.
> >
> > It also allows the model to be applied in a more step-by-step fashion
> > (there is no need to agree on that model worldwide).
> >
> > Any IX listening would like the idea?
>
> if anyone look into geo-addressing it would maybe not be the best to
> locate it around IXes. It is the setup we know best right now but what
> about thinking about further down the road? let the "tax-authorizies" or
> something similar in the countries/region do the work or something?
>

The basic point of the idea is that the address assignment is
associated to the point where the routing and de-aggregated routes are
exchanged, and multihoming transit peerings are made. Other
authorities in the region that do not do routing exchanges can't do
the work as efficiently.

> what I mean, it would be just simple to just let the IX do it (if they
> want to), but maybe it would be better to consider it in a bigger scope
> and some years down the line? Maybe IX would be the first out of many step
> towards geo-addressing...
>

Maybe. I think it would be a compromise between global hierarchical
geographical addressing and Internet decentralization (that today is
also based a lot in the IXs).


2007/7/21, John Payne <john at sackheads.org>:
> On Jul 21, 2007, at 10:30 AM, Roger Jorgensen <rogerj at jorgensen.no>
> wrote:
>
>
> Geo addressing sounds interesting on the surface, but every proposal
> seems to require a new economic model. I find it difficult to believe
> that will happen anytime soon.
>
I don't know, why wouldn't it work with the current peering model?

> The other problem with using IXs that immediately comes to mind is
> that you are essentially multihoming to a single piece of
> infrastructure. Not really resilient in my mind :)
>
I think that this only causes problems if all your multihoming
providers get disconnected from the IX...

At the level of traffic that they handle now, if a whole IX is brought
down, there will be a major disruption on your traffic anyway because
the international transit of your providers will not be able to cope
with all the new traffic.

And in many cases, most of the IXs do in fact have quite resilient
infrastructures (with even geographically redundant locations).

2007/7/22, Brian E Carpenter <brian.e.carpenter at gmail.com>:
> On 2007-07-20 14:17, Carlos Garcia Braschi wrote:
> > Those would act as LIR / RIR for all ISPs connected to them and would
> > have incentive to promote the idea... as it promotes fidelity of their
> > peering customers and gives them more service.
>
> What is the financial incentive on the various actors that will
> make this happen?
>
>      Brian

IX have more work to do and they can charge a little bit more. They
get a more optimized local routing, so they get more traffic.

The other actors still have the same incentive as the other
geographical addressing: more feasible multi-homing, which is not
much, but it's something...

2007/7/22, Nick Hilliard <nick-lists at netability.ie>:
> On Fri, Jul 20, 2007 at 01:17:54PM +0100, Carlos Garcia Braschi wrote:
> > I'm not sure if this has been proposed already, but what if we did
> > geographic addressing but assigned the addresses to
> > interconnect/peering points? (requiring that they be used to cover the
> > region the IX is in, and that among all the peering partners manage to
> > exchange locally the un-aggregated routing tables).
>
> Can we agree to consign the idea of geographic addressing to the scrap heap
> where it belongs, please?  It's not going to work, ever - and we need to
> stop pretending that it has any future.  It will work the day that the
> Internet (big "I") is operated as a strict tree structure, which is to say,
> never.
>

I think in that by anchoring the addresses to the IXs the internet
does not need to be operated as a tree structure, only in the same way
as today.

> Let's be clear here.  Why should a company operate its business using
> address space which is encumbered with shackles to other connectivity
> providers (IXPs, transit providers, etc)?  It makes no sense, because it
> deprives the company of the ability to make choices about their ability to
> route traffic on the Internet.  No sane business is going to put themselves

The reduction on choice has the advantage of more scalability of the
whole network and is more flexible than a purely geographical address
assignment. And allows customers better multihoming. Again, it could
be done only for multihoming addresses...

> in this position because this way lies monopolies, high prices and all the
> problems caused by lack of competition that we've historically seen.
>

There is already a monopoly on IPv4 and IPv6 addresses at the
wholesale level, with rather low prices.

> I have a better idea.  Instead of using IXPs who quite often have rules
> about not engaging in this sort of non IXP-related stuff, let's bring back
> the idea of the incumbent telco, and give them a monopoly on regional /
> national address space management.  And transit / connectivity too, because
> that's the only way to make geographic addressing really work.  And
> everything else.
>

Well, the main difference and advantage of IXPs is that they do not
provide transit, so do not have commercial interest in having a
transit traffic monopoly. And having them give the geographical
addresses does not prevent the addresses in private peerings, because
they do not have a monopoly on local peering.

But, if you are in an IX and are not convinced... can I have a second
opinion? :-)

> Is it now clearer why this is such a bad idea?
>
> Nick Hilliard
> --
> Network Ability Ltd. | Technical Operations    | Tel: +353 1 6169698
> 3 Westland Square    | INEX - Internet Neutral | Fax: +353 1 6041981
> Dublin 2, Ireland    | Exchange Association    | Email: nick at inex.ie
>