Re: [lir-wg] Discussion about RIPE-261

[Andre Oppermann <]

Nils Ketelsen wrote:
> 
> On Thu, May 29, 2003 at 11:36:16AM +0200, Andre Oppermann wrote:
> > Nils Ketelsen wrote:
> 
> > > So, what we really want is PI addresses. And with the current pratices they
> > > just do not aggregate which also is a bad thing. This is why I think the
> > > geographical approach already mentioned on the list (one netblock per
> > > country, different sizes depending on population) currently is the approach
> > > which fits that need best, I believe.
> >
> > How do you come to that conclusion? Every other PI space will be with
> > another ISP. So even thought you might have everything close together
> > on a "human logic" level there is no way to aggregate the prefixes
> > together in the routing system. Unless of course you want to do it
> > PTT style where one is the one who routes this block.
> 
> I think in most cases the "human logic level" also works for routing. So

No, it does not. And that is exactly the point. It never has and
probably never will as long as we don't have monopolies in each
country.

BTW have a look at how the country based E.164 telephony system is
implementing number portability...

> I guess many people will carry the complete routingtable for those countries
> they communicate a lot with (mostly this will be the own country and some
> nearby, but YMMV) and just have a few aggregates like "send all traffic to
> $farcountry1 to uplink A, send all for $farcountry2 via uplink B etc".

No. This could only hold true for maybe a leaf node (non-transit) with
two upstreams. Everyone else who is doing any kind of transit will
have to carry the full table. I ask you where are the savings?

Which network engineer wants to define for each and every $forcountry
where it goes?

BTW we have that already today. For example you can have a primary
uplink (default route) and a secondary uplink with a smaller ISP in
your country from where you only import his and his peer prefixes.
Under these circumstances even an obsolete 2500 can do BGP routing.

> Communication mostly happens in the same way "human logic" does, because in
> the end its humans initiating the communication.

To decouple exactly this human logic from the technical implementation
we have DNS and BGP. For example I'm in Switzerland. Interestingly the
website with most hits (blick.ch) is physically placed in Germany and
connected with a German ISP who does not even have a single line to
Switzerland. This is just one case where your "human logic" assumptions
breaks. And the Internet has exactly been designed to separate these
layers from each other. It should, it must not matter where your
resource is located.

> I think this could be a good compromise between every multihomed user has to
> be in every routing table and ongoing provider dependance.

Nope. It looks like you have never done BGP before. Until you have
got some significant experience, I'm sorry to say, you are not fully
qualified for discussion on this level. (What about me? Have a look
at AO6-RIPE, AS8271 and AS8235).

-- 
Andre