[ppml] geo addressing (was: Re: 2005-1 or its logical successor)

[Stephen Sprunk]

Thus spake "Christopher Morrow" <christopher.morrow at gmail.com>
> On 11/9/05, Stephen Sprunk <stephen at sprunk.org> wrote:
>> Thus spake "Howard, W. Lee" <Lee.Howard at stanleyassociates.com>
>> >> From: ppml-bounces at arin.net [mailto:ppml-bounces at arin.net] On
>> >> Behalf Of Stephen Sprunk
>
>> > I hear a lot of support for settlements from telcos,
>> > some from governments, and little from network people.
>> > It effectively drives hosting companies out of business,
>
> I'd be concerned, am concerned, about 'settlement' situations. Perhaps
> its a misunderstanding of them on my part though :) I'd be concerned
> that there are more than 2 parties involved in the conversation, and
> potentially someone in the middle could bend traffic paths to
> influence their business needs (putting competition out of business or
> just censoring people?) Anyway, probably not a topic for this
> conversation, just worrisome :) (knowing how evil bean counters can
> be...)
>
>> > if they have to pay for transit (settlement for inbound
>> > traffic) but carriers get to charge at both ends of the
>> > stream.
>>
>> OTOH, people with big upstream pipes, like hosters, could end up
>> receiving huge settlement checks for inbound traffic on pipes that
>> are today only filled in one direction.  I've seen no data to indicate
>> that the net
>
> If they charge lots of settlement, perhaps people won't send packets
> their way? or will degrade performance intentionally? Or force that
> traffic through a competitor to impale them on a higher cost link :(

Since all IX members would be advertising the same aggregate to their 
upstreams, inbound traffic necessarily flows to the member closest to the 
source.  All one can do is _prevent_ traffic from reaching you by not 
advertising, prepending, etc. but why would they do that when they stand to 
make money for every bit of yours that comes in and avoid paying you for 
every bit of theirs they get directly?

>> > Is there evidence that this model provides a network
>> > that is better, faster, or cheaper?
>>
>> It's arguably "better" and "faster" because local traffic stays local and
>> inbound traffic takes the most direct route; at a macro level, both are 
>> good
>> for the Internet.  Less long-haul traffic means transit prices should 
>> drop,
>> leading to "cheaper".
>
> it takes the most direct route assuming no funny monkey business with
> routing... assuming people don't have capacity problems in region (or
> at the IX) and don't force traffic on non-optimal paths.

Assuming the settlement rate is comparable to or higher than the current 
transit rate (something nobody has addressed so far), everyone has a 
financial incentive to build as much inbound capacity and attract as much 
inbound traffic as possible.

>> Also, by replacing PI-based multihoming with IX-based multihoming,
>> there is less pressure on the DFZ, leading to cheaper routers, or at
>> least
>
> I am not sure that local/regional IXP proliferation is going to help
> the DFZ folks all that much in the end. It may remove direct
> connections and direct allocations from their tables. It would
> essentially aggregate all customers in region behind one prefix.
> Eventually the 'regional' IXP would have to become a County IXP then a
> City IXP then Town IXP, proliferating the number of IXP's to a very
> large number (how many towns exist in the USA alone? Does this really
> drop the DFZ table that much?

There are ~3000 counties in the US; while all of them certainly don't need 
their own IX, even if we did something that nutty 3000 routes for the US is 
a clear improvement over the status quo.  If we restricted "areas" to, say, 
the so-called NFL cities or something similar, we'd have another two orders 
of magnitude reduction in routes.

There's certainly the possibility of IXes above the first level, but I 
haven't thought that much about how it would be accomplished.  Frankly, I 
think we'd be okay with 3000 routes for the US, and I'm not sure that 
getting it down to 1 route is worth the extra hassle.

> Additionally, the DFZ folks now become the 'LD carrier' of the
> Internet and since they don't have direct customers they arrange
> 'settlement' with the IXPs I suppose? I'm not sure this gets us to the
> end goal either.

The way I've seen it described, there's nothing preventing any ISP from 
selling transit to any other or to end sites; the only thing is that to use 
the IX-based addresses in a given area, they'd need to peer at the IX (and 
join the settlement scheme).

So, your promising local ISP could join the IX, sell a customer a 
connection, buy transit from UUNET et al, and nothing much would change 
except that (a) your customer could multihome (or rehome) to any other IX 
member without any new routes in the DFZ, (b) you would get settlement check 
for traffic coming in on your transit pipes headed for other IX members, and 
(c) you'd send settlement checks for traffic coming in on other IX members' 
transit pipes headed for your customers.

I'd originally looked at this without settlements, thinking that the 
benefits to all would provide the motivation.  Unfortunately, that _does_ 
lead to the inbound traffic problems you describe.  Ironically, with 
settlements I think everyone will build to the point their settlement 
revenue/expense will be roughly zero -- and as a side effect build a more 
efficient Internet for everyone.

>> less-frequent upgrades.  And multihoming, at least within one "area",
>> becomes significantly cheaper and easier, which a lot of end sites would
>> call "better".
>
> How is a local IXP 'multihoming' me? it's getting me a connection to a
> single 'carrier' with lots of bgp options... or I'm again confused. It
> may get me more than one pipe to the IXP and more bgp options on each,
> and now the IXP-truck-bomb is my failure scenario (and much easier I'd
> think).

As I said, there are (ugly) ways of handling failure of a single IX.  The 
better solution is to have multiple IXen per "area".  Frankly, in most fiber 
hub cities there's a handful of buildings to bomb and you could take down 
most of the Internet (heck, the phone network) within hundreds of miles. 
Competent folks would look at how much redundancy is logical for a given 
physical plant and build to that level -- and no more.

S

Stephen Sprunk        "Stupid people surround themselves with smart
CCIE #3723           people.  Smart people surround themselves with
K5SSS         smart people who disagree with them."  --Aaron Sorkin