[ppml] 2005-1 or its logical successor

Stephen Sprunk stephen at sprunk.org
Wed Nov 9 15:38:25 EST 2005


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
>
>> True, but in the various discussions I've seen of this idea,
>> the natural conclusion seems to be that each geographical
>> area would have an IX that all ISPs using the PI block would
>> be required to connect to, and each ISP would advertise
>> only the aggregate to their transit providers.  It appears
>> settlements would be required for traffic coming in on one
>> ISP's links and headed to another ISP's customers;
>> upstream traffic would be handled as it is today.
>
> What's the transition plan, if these IXs don't exist now?

Interesting question, and one of the stumbling blocks to getting this 
approach deployed.

> Are these IXs required by law, by policy, or something else?
> Are they privately-owned monopolies, or publicly-owned
> monopolies?

They're effectively required at a technical level, though I don't see them 
getting created without government regulation.  In the absence of the 
latter, I'd suspect they'd be set up as non-profits owned by their members 
or something similar; if the gov't got involved, I'd expect them to be 
private for-profit companies selected by bribed politicians.

>> This model effectively trades a BGP routing problem for a
>> money routing problem.  Given no significant improvements
>> have been made to BGP for a long time, perhaps it's time
>> to let the bean-counters have their shot?
>
> 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,
> 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 
movement of money would be substantially different under this plan (or that 
it wouldn't, for that matter).

> 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".

Also, by replacing PI-based multihoming with IX-based multihoming, there is 
less pressure on the DFZ, leading to cheaper routers, or at least 
less-frequent upgrades.  And multihoming, at least within one "area", 
becomes significantly cheaper and easier, which a lot of end sites would 
call "better".

However, I don't think any ISPs in isolation would see enough "better, 
faster, cheaper" to justify making the move.  It only works if everyone does 
it, which is contrary to human nature.  We're long past the days of doing 
things for the common good, at least when they require a fundamental change 
in business models.

> If economists set addressing policy, can network engineers
> set monetary policy?

Non-sequitur.

>> I'm also wondering how many "tier 1" providers would be willing to
>> participate in such a model absent government regulations.
>> Why would UUNET want to do something that makes it easier
>> for their customers to multihome (or rehome) to "tier 2/3"
>> providers or even another "tier 1"?   Would we see
>> significant benefits with just the smaller ISPs participating?
>
> Because those providers are inferior.  The phrase used
> to be, "What's good for the Internet is good for UUNET."
> That's tongue-in-cheek, of course.  However, the same
> company owns UUNET and MAE-*, so there might be some
> interest,

I don't buy that argument.  I can't think of a reason that an IX-based model 
would improve a "tier 1's" profitability, and can think of several reasons 
they wouldn't.  At best, I see them grudgingly joining the club after all of 
the "tier 2/3" players form a united front and use it as a marketing tactic.

And owning an IX here or there isn't meaningful when we're talking about an 
order of magnitude (or two) growth in the number of IXen -- unless one 
manages to get a profitable contract operating a number of them.

> but I think having a single point of failure (nuke the IX) in
> each geographical area sounds contrary to good
> internetwork design.

Dillon points out that an IX isn't strictly required; for small enough 
numbers of participants, you could create a full mesh of private 
connections.  You could also do this as a backup for the IX.  If the mesh 
isn't full (due to circuit failures, most likely, or due to IX failure), the 
partially-connected party(ies) would have to leak more-specifics upstream 
and withdraw the aggregate, but obviously this degenerates to today's PI 
situation (and possibly worse) if not quickly corrected.  Global impact of a 
local problem is a Bad Thing(tm).

All in all, I'm not sure I support this idea.  It's technically interesting, 
but I'm not sure if it's viable in today's market.  If we'd done it a decade 
ago, however...

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 




More information about the ARIN-PPML mailing list