On a future of open settlement free peering

Richard Bennett richard at bennett.com
Tue Jul 29 22:21:32 UTC 2014

It's interesting that an FCC ban on paid peering (or "on-net transit" if 
you prefer that expression) is now seen as a plausible and even likely 
outcome of the FCC's net neutrality expedition. It wasn't that long ago 
that a number of NANOGers insisted that such action by the FCC was 
totally out of the question and could only be suggested by clueless 
industry shills. I'm talking about a blog post I wrote for GigaOM in 
2009 about how video streaming was changing the Internet and the FCC's 
curiosity about paid peering raised in Question 106 of the 2010 Open 
Internet NPRM, in which I wrote:

"But paid peering may be forbidden by Question 106 
<http://hraunfoss.fcc.gov/edocs_public/attachmatch/FCC-09-93A1.pdf> of 
the FCC’s proposed Open Internet rules because it’s essentially 
two-tiered network access, [Bill] Norton points out.

" Paid peering illustrates how hard it is to write an 
anti-discrimination rule for the Internet that doesn’t have harmful side 
effects for all but the largest content networks. Paid peering is a 
better level of access to an ISP’s customers for a fee, but the fee is 
less than the price of generic access to the ISP via a transit network. 
The practice of paid peering also reduces the load on the Internet core, 
so what’s not to like? Paid peering agreements should be offered for 
sale on a non-discriminatory basis, but they certainly shouldn’t be banned."


In the comments, Daniel Golding and a number of others who hung out on a 
particular IRC channel insisted an FCC ban on paid peering was totally 
out of the question: " Paid peering is not and will not be banned. Bill, 
unfortunately, made this up. There is no way to read the proposed 
rulemaking this way – its simply not in the document, at all."

One of the more interesting wrinkles was Patrick Gillmore walking back 
from things he and his co-authors had said about paid peering in an 
academic paper a few years earlier, such as:

“We also have a cautionary conclusion: if one should be motivated (for 
whatever reason) to contemplate some regulatory rule to manage 
interconnection (which the debate over Net Neutrality is, in part, 
about), the design of such a rule will be both complex and 
informationally demanding. Any simplistic rules that try to define 
network neutrality as the elimination of discrimination will fail even 
to match today’s reality by a wide margin. There is a substantial level 
of economic discrimination today just in the variation in willingness to 
peer, and the emergence of paid peering and partial transit only 
increase this space. Partial transit and paid peering may be seen as 
efficiency-enhancing responses to changing market conditions. While 
there may be opportunities for abuse by providers with excessive 
bargaining power, the complexity of what is in place today, and what 
seems to be working today, would argue that the best way to address any 
potential concern would be to focus on the sources of bargaining power 
and identify anti-competitive opportunism, rather than to impose ex ante 
restrictions on the range of bilateral contracts.” – Complexity of 
Internet Interconnections: Technology, Incentives and Implications for 
Policy, P. Faratin, D. Clark, P. Gilmore, S. Bauer, A. Berger and W. 
Lehr. http://people.csail.mit.edu/wlehr/Lehr-Papers_files/Clark Lehr 
Faratin Complexity Interconnection TPRC 2007.pdf

It seems that those of us who predicted FCC involvement in peering in 
the name of net neutrality were right, regardless of the terms we used, 
the sources we cited, or where our paychecks came from at the time.

If you read the initial net neutrality arguments from Tim Wu, Larry 
Lessig, and Mark Lemley, it's pretty clear that net neutrality was 
always about interconnection. In "Network Neutrality, Broadband 
Discrimination" Wu discussed it in terms of a "gateway" between a 
broadband network and "the Internet"; see: 
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=388863 .

That was obviously a simplification, but it's clear that his "gateway" 
represented the entire interconnection apparatus connecting a given 
eyeball network to the rest of the Internet. And perhaps it's also clear 
that this "gateway" can never be neutral because it comes down to ports 
at specific locations and a distribution and aggregation fabric between 
residential connections and ports that can never provide exactly the 
same service on every possible path.

 From one point of view, questions about fair and equitable 
interconnection have to do simply with fees and capacity, but from 
another they have to do with steps the traffic source should take to 
ensure that it releases traffic into the eyeball network at locations 
that permit the most efficient internal routing.

And other questions are beginning to emerge about the poor interaction 
of video rate control algorithms and TCP congestion control that 
indicate that services such as YouTube and Netflix are actually their 
own worst enemies, see: 
http://apps.fcc.gov/ecfs/document/view?id=7521706465 , page 13, and 
http://apps.fcc.gov/ecfs/document/view?id=7521389953 , page 8.

So yeah, the demand for "free and open" interconnection is front and 
center, and it tends to submerge questions about the obligations of 
traffic sources to deliver to the best locations in an efficient way. 
There certainly are opportunities for abuse on both sides of the "gateway".


On 7/29/14, 10:30 AM, William Herrin wrote:
> Howdy folks,
> It seems to me that we're moving in a direction where either
> ratioless, high-capacity settlement-free peering will be a industry
> requirement exercised voluntarily, or where some heavy-handed
> government regulation will compel some kind of interconnection that
> the holdouts find even less desirable. I can only hope the holdouts
> will "see the light" before the weight of government crashes down on
> them -- regulation has no winners, only losers and bigger losers. And
> sometimes the worst thing that can happen is you get what you ask for
> with no opportunity to later change your mind.
> I'm curious what lies beyond that horizon. If we stipulate for the
> sake of the discussion that open peering is the way it going to be, a
> critical part of network neutrality, what exactly will that mean?
> Will it be permissible for one network to ask the other to pay a
> one-time port cost for the initial interconnect, assuming its
> representative of the actual cost of a one-time equipment addition?
> To what degree is redundancy a requirement? If a network refuses to
> peer in more than one chancy location, does that mean their peering
> policy isn't really open?
> Will a network be compliant if the open peering connections are only
> available in its own data center? Or will they need to be available in
> neutral data centers?
> Would a refusal to connect to neutral peering fabrics constitute a
> refusal to connect to smaller networks? Or is it reasonable to state
> that anybody who can't come up with 10 gig ports and cross-connects
> isn't of threshold size?
> Can a peering policy be open if it's regionally restricted? If my
> peering points for the mid-Atlantic states only announce routes tied
> to my mid-Atlantic customers and only propagate your routes to those
> mid-Atlantic customers, is that acceptable behavior? Or have I
> mis-served my customers if I don't pull all of them to the location
> you find it convenient to peer?
> Food for thought,
> Bill Herrin

Richard Bennett
Visiting Fellow, American Enterprise Institute
Center for Internet, Communications, and Technology Policy
Editor, High Tech Forum

More information about the NANOG mailing list