Friday, November 14, 2008

Tim Lee's Reasonable Retorts

Tim Lee has posted two responses to my critique of his recent article on network neutrality. In his article, he argued that network neutrality is good but that violations of the principle are unlikely to be severe for economic and technical reasons. I find this argument unpersuasive, and he responded with four basic retorts:

1. Net Neutrality proponents don't clearly state what they are seeking to prevent, and thus evade any attempt to disprove their harms.

I’ve found that any time I take one of these ISP strategies seriously and put forth an argument about why it’s unlikely to be feasible or profitable, the response from supporters of regulation is often to concede that the particular scenario I’ve chosen is not realistic...

Lee goes on to list several scenarios, all of which are possible to varying degrees. However, they all fit the simple rubric of network discrimination and they all are harmful. In general, subtle discrimination is more likely than outright blocking. This is something that has been clearly articulated by the mainstream of neutrality proponents for some time. That is why I included reference to Barbara van Schewick's paper. If Tim were choosing to "take one of these ISP strategies seriously" he would have done well to focus on the one that most people are talking about.

2. This type of discrimination is unlikely, isn't that bad, and we can always fix it after the fact.

First, notice that the kind of discrimination he’s describing here is much more modest than the scenarios commonly described by network neutrality activists. Under the scenario he’s describing, all current Internet applications will continue to work for the foreseeable future, and any new Internet applications that can work with current levels of bandwidth will work just fine. If this is how things are going to play out, we’ll have plenty of time to debate what to do about it after the fact.

I am describing a mainstream version of discrimination, which can happen either right now or going forward as operators upgrade their networks but keep non-payers in the slow lane. We have ample evidence of the former in Comcast/BitTorrent. The latter is simply a less visible version of the former -- an even further degree away from Lee's scenario in which consumers have "a taste of freedom", "become acutely aware of any new restrictions," and, "stubbornly refuse efforts to impose them." The fact that carriers are building out faster networks doesn't tell us whether or not this is likely. Carriers will of course build out faster networks, because they typically profit more from them (whether they impose discrimination or not). Given the current uncertain regulatory climate, it is no surprise that they have refrained from additional large-scale discrimination. This climate, however, is temporary. The relevant question is whether or not those network upgrades provide additional shield from the customer backlash that Lee posits. It is clear that they do.

How bad you think this discrimination is depends on how seriously you take arguments about platform economies, dynamic innovation, network effects, and freedom of speech. It also depends on whether or not you think that degrading service achieves most of the ends of outright blocking. I argue that it does. Google's obsession with page load times is not simply because they are hyper-focused engineers. Skype's need for equal network treatment is not just because they want calls to sound nice. The BitTorrent protocol's expectation that connections are not randomly reset is not a matter of convenience.

Lee would have us believe that we will always have the space to regulate these issues, if needed, after the fact. The Comcast order might give us some hope in this regard, except for the tremendous murkiness that surrounds the decision, its implications, and its legal durability. Regulation from the FCC can be roughly thought to fall into two categories: rulemaking and adjudication. Rulemaking explicitly sets out the detailed requirements, whereas adjudication defines basic guidelines and then builds policy through case-by-case enforcement. Lee clearly opposes rulemaking on its face. We are left with adjudication, but in this case he opposes further definition of enforceable principles. This is not ex post regulation, it is no regulation at all.

3. The risks are overblown, and disproved by history.

It’s worth remembering that alarmism about the future of the Web is almost as old as the Web itself.

Lee is not a fan of Lessig's "apocalyptic" predictions in 1999. While Lessig's forecasts undoubtedly have not fully come true ("yet" -- as he notes in the preface to the new edition), we have unquestionably seen some of those trends play out. Increasing control by intermediaries, domestically and abroad, threatens speech and innovation. The "open access" battle that was heating up at that time was not lost until 2005, and since then we have case studies for how the stopgap quasai-neutrality principles are strained. But, I'm not here to defend Lessig (I certainly disagree strongly with him at times).

Rather than debating generally whether past predictions of others have come true, it is more productive to examine the specific issues at hand with the most relevant data points from history and the present. We know that historically corporations tended toward building closed systems like AOL and CompuServe. We know that well-crafted regulatory interventions like common carrier non-discrimination, Computer II, and Carterphone unleashed waves of innovation. We know that carriers today have pursued discriminatory practices and been partially disciplined by somewhat ambiguous regulation. We know that abroad, discriminatory practices have flourished in environments in which intermediaries exercise the most control. We know that domestically in the parallel (and increasingly overlapping) wireless market, market actors impose restrictions that radically limit innovation.

This is not a strong historical or factual case against the need for, or success of, non-discrimination regulation.

4. Steve misunderstands settlement-free peering.

“Settlement-free” means that no money exchanges hands. If D and E are peers [this example assumes that D is a "last mile" backbone provider like Verizon and E and F are competitive "tier 1" providers such as Level 3 or Global Crossing], that by definition means that E pays D nothing to carry its traffic, and vice versa.

This technical/wonky definition is at the heart of what I consider Lee's most original, but nevertheless misguided, argument. The basic idea he posits is that because a certain set of backbone providers traditionally negotiate no-fee interconnection agreements, there is no ability for last-mile providers to leverage their power in the consumer market into the backbone market.

Let's go back and define a couple of key terms. First, "settlement-free peering" means, as Lee accurately describes, an arrangement between two providers in which they do not exchange money but simply agree to carry each others' traffic. They do so under detailed and confidential interconnection agreements that define the terms of this agreement, including things like jitter, latency, throughput, etc. These agreements often require equal treatment by both parties (although they may not speak to those providers' relationships with other providers). Let's assume for the sake of argument that they always do require equal treatment between the two. The types of companies that have these agreements are "Tier 1" backbone providers at the core of the internet -- Level 3, Sprint, AT&T, etc.

Second, "transit" agreements are contractual relationships between unequals. In this case, one party typically pays the other for carrying its traffic under various terms. This is the type of relationship that Comcast has with the Tier 1 providers. For example, here is an excerpt of the traceroute from my Comcast cable modem to

7 (
8 (
9 (
10 (

See that? My packets go from Comcast -> L3 -> Google. Comcast pays Level 3 to transmit their packets, according to some confidential terms that it agrees to. Comcast has a rather large national network (although it is not a "Tier 1" provider) and thus can route its packets around to locations where it has the best bargaining power with the party at the exchange point (in this case, they sent my packets from Boston to Chicago before plugging into L3). Lee's theory is that the settlement-free peering agreements probably don't allow discrimination based on content or source, and he seems to assume that downstream transit agreements are implicated in this obligation because at some point they must interconnect with those backbone providers. Furthermore, he claims that both parties need each other enough that nobody would ever violate these principles.

In my initial critique, I gave several reasons to doubt this claim. First, there is no practical evidence that Tier 1 providers have pressured their downstream transit peers to remain non-discriminatory. This has not been a factor in discrimination disputes that we have seen to date, like Comcast/BitTorrent or Madison River (instead, regulatory threats have brought players in line). Second, there is ample reason to believe that Tier 1 providers would indeed be willing to de-peer despite Lee's assertion that they simply need each other too much (thus I cite the Cogent/L3 dispute as well as the Cogent/Sprint de-peering from a couple of weeks ago). Third, the universe of settlement-free peering is increasingly giving way to varieties of transit agreements in which concessions are made in exchange for payment. Fourth, there are now emerging unified backbone/last-mile networks for which much of the traffic need not pass through a Tier-1 exchange point at all (eg. Verizon/MCI/UUNET). Settlement-free peering has been a powerful norm in keeping content or source-based discrimination out of the core of the network, but even there their strength is waning.

No comments: