Re: geo short vs long term? [Re: Geo pros and cons]

[Sean Doran <smd@ab.use.net>]

On Sunday, Apr 6, 2003, at 10:27 Europe/London, Iljitsch van Beijnum 
wrote:

> The idea behind geographic addressing is not that the topology and 
> addressing become interchangable. The simple fact that a multihomer 
> connects to the net in two places makes this impossible by definition.
No, this is not true, unless you insist a multihomer only ever uses one 
address
to fit a constrained L2 topology.

Multi-address-with-existing-v6-semantics and 
multi-address-using-GSE-or-other-split-I/L
approaches both preserve the direct relationship between topology and 
address or locator.

MPLS/2547 and other line, cloud and lan virtualizations ("liberate your 
Internet traffic from the
tyranny of physical infrastructure!") likewise, only the other way 
around: they can adjust topology
to fit existing addresses.

> The point is that it becomes possible to draw lines on the map in such 
> a way that aggregating routing information that crosses these lines 
> gets rid of enough routing information that the savings in routing 
> table size are worth the effort.
>
> For this purpose, it is irrelevant that the aggregation circle with 
> Singapore in the middle may also include Palo Alto and LA. That still 
> gets rid of Asia/Pacific more specifics in most of the US and the rest 
> of the world. And even if some Asian networks connect to other places, 
> this only breaks aggregation for these specific networks.
>
> Maybe the savings aren't that big.
The savings are enormous.  Aggregating the entire western hemisphere 
behind
a single prefix would be wonderful.  (I propose using existing Sprint 
address space.)

However you are overlooking the fact that there are costs too (on top 
of the
monopoly rent I will gladly extract from hundreds of millions of people 
like you,
who will deliver all your North American traffic only to me).

In particular, should connectivity between me and something covered by 
my single
prefix fail, you and everyone else in the eastern hemisphere will not 
know this -- not
a really problem, except for the cases where there are longer paths 
available which still work.

I suggest that aggressive regulation will constrain this problem, so 
engineers
need not think about the dissolution of abstraction boundaries, and 
other things
that go bump in the night.

Massive spending on local redundancy and resiliency is the correct 
answer.
The industry's approach -- connecting to more than one provider -- is 
fundamentally
a bad idea.  Fortunately, gentle guidance will resolve this surprising 
failure of the market,
and lead to a more stable and reliable Internet for everyone.

> But then again, the effort isn't all that huge either: the RIRs need 
> to implement a tool that allows local internet registries (ISPs) to 
> give out geography-based /48s to multihomers. That's all.
My approach is even easier.  "No, you can't have an address, unless you 
get it from Sprint".
This requires no new tools (they have already deployed the vacation 
program and procmail),
and offers ARIN a considerable opportunity for staff reduction.

I also think it's cheaper to pay me obscene one-provider charges in a 
few years than it
will be to deal with a routing system with a hundred thousand 
residentially multihoming
/48s in each of hundreds of major cities.   My neighbours, oddly 
enough, use different
providers than I do, so their providers would have to know about my 
/48, and mine theirs,
and no aggregation is possible for this immediate area for any set of 
providers operating
within the closest abstraction/aggregation boundary.   This is not 
atypical, in my experience,
and boundary pressure is likely to force a trade-off between leaking 
global exceptions and
increasing local state.

Sigh.  I wish I were the One True Monopoly.   You should too: there'd 
be less of this sort of email.
Try working on SAPI: "Sean-Assigned Public Internet addresses".    
That'd be much cooler.

	Sean.