AOL & Cogent

Paul Vixie paul at
Sun Dec 29 22:32:25 UTC 2002

> > ... if everybody who could peer in N places worldwide could just get
> > peering, then all kinds of per-bit revenue for "high tier" network
> > owners would turn into per-port revenue for exchange point
> > operators. ...

> Well, I think as a local operator you can not expect to be able to
> peer with everyone to receive global routes but theres no reason not
> to exchange local routes comparable to the area your own network
> covers, this wont affect transit sales and wont cost you in backhaul
> either. Thats a slightly different perspective than assuming you can
> get a providers to exchange all their network with you in a settlement
> free bilateral.

there we go again, talking technology and making the technological kind
of sense.  peering isn't a technology decision, it's a business decision.

as a local operator myself (ISC), i know that i should not expect peering
other than if someone wants their customers to have better access to the
f-root server or the ftp server or whatever.  it's actually
easier for me, as a nonprofit, to attract what mr. bill calls 'content
peering' relationships, since i don't compete with the folks i peer with.

however, in a former job, i took the reins of abovenet and used a lot of
mfn fiber and mfn resources (back when they had resources to use, that is)
and built a network that touched down in more places with more gigawhuppas
and more bit-miles and so on than about half of the current so called "top
tier" networks had then (or indeed have now).  i surrounded PSI on all sides,
with a network that carried more actual traffic and had more provable
headroom, and more endpoints.  yet they still insisted on playing peering
games.  (perhaps if they'd won those games they would still exist today?)

peering is not about equity, or ratios, or technology.  it's about money.
sadly, too many people are focusing on their share of the current market
rather than on the size of the eventual market, so, short-term thinking
pervades the space, and the actual customers who source and sink all this
traffic don't ride a curve that looks anything like moore's law.

try a thought experiment.  take about $450M in vee cee money, buy up a lot
of bankrupt capital and routes, hire a bunch of starving backbone engineers
and sales/marketing/finance/etc people at downsized salaries, and build a
network that attends about 40 major carrier interconnect locations (some
internet exchanges, some carrier motels).  document the hell out of it, so
that when you enter peering negotiations with the current "top tier" networks
you've got "attachment A" already done and audited.  now ask yourself the
chances of becoming defaultless and settlement free before you run out of
cash.  (now, does anybody still think peering is a technology issue?)

> And definitely to your gamers and possibly your VoIP folks to
> (depending on details) they will be very fickle on your network
> connectivity and the quality of local peerings is crucial to these
> applications, gaming is growing very quickly as more people get flat
> fee broadband and to a residential access provider I wouldnt
> underestimate how much it could hurt to increase the path to the
> servers by a couple of hops.

like i said, we're living in the shadow of bankrupt overcapacity, and until
we burn it off, "cost per bit-mile" is going to be too low to measure when
compared with "cost per peering edge".  the next six to twelve months will
favour the "small number of large peering edges" model.  once the capital
and routes are rightpriced, and transit contracts are rightpriced, and we
reach some kind of equilibrium between the value and cost of traffic, then
some kind of technological argument about peering might hold some sway.

More information about the NANOG mailing list