The last decade’s worth of US policy work on broadband Internet openness – first open access, now net neutrality – has focused largely on the access links operated by individual residential ISPs. But the recent dispute between Level 3 and Comcast has served as a jolting reminder that the Internet is a network of networks that only functions because those networks agree to interconnect with each other. As that agreement turns to disagreement, it reveals just how strained the Internet interconnection market is becoming.
As the dispute has unfolded, opinions have proliferated as to which company is in the wrong and who is exerting market power over whom. Without knowing all of the commercial details, it is difficult to interpret each company’s claims. But a vague picture of the situation is beginning to emerge.
Comcast has been a transit customer of Level 3 for a number of years (“transit” and other relevant terms are defined below). With 17 million subscribers, recently upgraded residential broadband capacity, and one of the Internet’s largest shares of inter-domain traffic, the size of Comcast’s network has allowed it to peer with many other large networks, but it still requires some transit to provide access to the full Internet, which is why it buys access from Level 3, Tata, Global Crossing, and perhaps other transit providers.
At times, Comcast has sought to peer with certain networks, and those networks have refused, as often happens in the process of negotiating such arrangements. Some observers have claimed that when networks refuse to peer with Comcast, Comcast routes their traffic over its Tata transit connection, which Comcast purposefully runs in a congested state so as to induce those networks to become peers.
Comcast has also been charging content delivery networks (CDNs) like Akamai and Limelight to interconnect (and, in some cases, to colocate their CDN servers in Comcast’s facilities). This provides better performance for the CDN content since it need not traverse transit links to reach Comcast’s users. It is unclear whether Verizon, AT&T, and other residential ISPs charge CDN companies similar rates to what Comcast charges for interconnection (Level 3 claims that Comcast’s charges are higher ). But Comcast is certainly not alone in charging fees to CDNs. Verizon, for example, charges CDNs for direct interconnection at regional hubs under its Partner Port Program.
Netflix, which by one account has been driving 20% of downstream Internet traffic in North America at peak times, recently shifted a substantial portion of its content from Akamai to Level 3 (although it retains commercial relationships with both Akamai and Limelight). This has led many observers to speculate as follows: after sealing the deal with Netflix, Level 3 informed Comcast that its interconnecting traffic would be increasing by more than a factor of two; realizing that this increase would be caused by Netflix content hosted on Level 3’s CDN, Comcast informed Level 3 that it would be charging Level 3 as a CDN. Level 3, which may have won Netflix’s business by, among other things, charging Netflix a lower price than competitors could afford on the assumption that it would not be paying Comcast to deliver Netflix traffic to Comcast’s subscribers, was angered by this and decided to make the dispute public and call for regulatory intervention.
Level 3’s claim is that Comcast’s actions violate net neutrality. It is difficult to see how Comcast’s demands could be construed as discriminating against particular content, applications, or services on the wire or in the routing infrastructure (especially if Comcast’s claim that Level 3 never mentioned Netflix during their negotiations is true ). The arrangement between the two companies seems to be based on quantities of traffic, not traffic content. So it is unlikely that this dispute raises net neutrality concerns of the sort that the FCC has been grappling with over the past several years. But whether Comcast designs its CDN pricing structure based on what traffic it thinks the interconnecting CDNs are hosting is certainly a valid question. Netflix was competing with Comcast’s TV and video-on-demand services before it moved some of its business from Akamai to Level 3. There was no public dispute about discrimination at that time, but that does not mean it was not occurring.
Might there be other reasons for regulatory intervention (even, as Level 3 has suggested, in the context of the Comcast-NBC Universal merger review)? Learning more of the facts would be a useful first step and one that regulators are uniquely positioned to help with. How much was Comcast paying Level 3 under the transit agreement? Were the two networks exchanging peering traffic on a settlement-free basis? How much do Comcast’s competitors charge for CDN interconnection? Were Comcast’s demands unreasonable by comparison? Does Comcast purposefully run its Tata link congested so as to convince networks to enter into peering relationships?
The answers to these questions are needed before passing judgment about who is in the wrong and who should be paying whom. The Comcast customer base creates significant demand for online content, including movie and TV streams from Netflix. By offering its service, Netflix creates significantly increased demand for the content it distributes. In accordance with the long-standing Internet model where every end user pays his own way, Netflix and customers of Comcast are both already paying their respective providers, Level 3 and Comcast, for Internet access. But when it comes to which of those providers should bear the brunt of the cost of the increased traffic demand, it is not immediately clear whether that cost should fall entirely on one side or the other, or be shared between the two.
That there is a cost associated with increased traffic demand should not be in doubt, however. Although the price of bandwidth is falling, the potential for increased demand created by services like Netflix streaming is massive. Comcast has indicated that Level 3 requested 27 additional ports as part of the negotiation, suggesting a 270 gigabit per second peak capacity requirement for the Netflix traffic. Such an increase added to current peak-hour traffic could easily come with a multi-million dollar price tag for backhaul capacity alone.
Who should bear this cost? The answer depends in part on one’s characterization of the competitiveness of the residential broadband market. Some observers have claimed that Comcast is able to strong-arm Level 3 into paying for interconnection because Comcast serves as the gatekeeper to all its own customers. But that is the case with every residential ISP – they all have terminating monopolies. Verizon, AT&T, and other ISPs are not having the same disputes with Level 3 (at least not in public), which means that there is something unique in the Comcast situation. It may well be the size of the Comcast network, which allows it to gain more favorable terms over comparatively smaller CDN businesses like that of Level 3. It may also be the case that, as the nation’s largest ISP, Comcast need not worry about losing customers if the performance of some services degrade while they sit behind the supposedly congested Tata link, whereas the other ISPs are in a more precarious position when it comes to customer retention. But that would indicate that there actually is residential ISP competition, not the opposite.
Shedding some additional light on the traditionally murky world of interconnection agreements would go a long way towards clarifying some of these ambiguities. But as regulators venture into this space, they should be cognizant of the global nature of these issues. The market for Internet interconnection has undergone rapid change in recent years and the dispute between Level 3 and Comcast is representative of the kinds of strains that have crept in to interconnection agreements among large network operators in countries all over the world. While previous policy issues surrounding broadband Internet openness may have been insulated from the Internet’s global dimension, venturing into the agreements that govern the network of networks will undoubtedly take the policy debate across national borders.
(see Complexity of Internet Interconnections: Technology, Incentives and Implications for Policy for an excellent primer on these topics)
• Transit – An interconnection arrangement in which a smaller networks pays a larger networks for access to the rest of the Internet.
• Peering (also known as settlement-free peering) – An interconnection arrangement in which two similar-sized networks agree to exchange traffic only with each other for no fee. For agreements between very large networks, the peers are usually expected to keep the ratio of traffic flowing in each direction to around 2:1.
• Paid peering (also known as settlement-based peering) – Like peering, except one party pays the other, usually because the traffic ratio between the two networks is asymmetric.
• Tier 1 network – A networks that reaches the entire rest of the Internet solely through peering. The Tier 1 networks peer only with each other and do not pay any transit or paid peering fees. Examples include Level 3, Global Crossing and Sprint.
• Content delivery network (CDN) – A network that distributes content by locating it close to end users. Exactly how close the content is located to users depends on the interconnection arrangements between CDNs and residential ISPs. Some CDNs pay to interconnect with ISPs. Examples of CDNs include Akamai and Limelight.