January 2008

One of the things I disagreed with in Yoo’s paper is that he puts a lot of stock in the notion that Akamai is a violation of network neutrality. Akamai is a distributed caching network that speeds the delivery of popular content by keeping copies of it at various points around the ‘net so that there’s likely to be a cache near any given end user. Yoo says that the existence of Akamai “attests to the extent to which the Internet is already far from ‘neutral.'” I think this is either an uncharitable interpretation of the pro-regulation position or a misunderstanding of how Akamai works.

Network neutrality is about the routing of packets. A network is neutral if it faithfully transmits information from one end of the network to the other and doesn’t discriminate among packets based on their contents. Neutrality is, in other words, about the behavior of the routers that move packets around the network. It has nothing to do with the behavior of servers at the edges of the network because they don’t route anyone’s packets.

Now, Yoo thinks content delivery networks like Akamai violate network neutrality:

When a last-mile network receives a query for content stored on a content delivery network, instead of blindly directing that request to the designated URL, the content delivery network may redirect the request to a particular cache that is more closely located or less congested. In the process, it can minimize delay and congestion costs by taking into account the topological proximity of each server, the load on each server, and the relative congestion of different portions of the network. In this manner, content delivery networks can dynamically manage network traffic in a way that can minimize transmission costs, congestion costs, and latency…

The problem is that content delivery networks violate network neutrality. Not only does URL redirection violate the end-to-end argument by introducing intelligence into the core of the network; the fact that content delivery networks are commercial entities means that their benefits are available only to those entities willing to pay for their services.

I think Yoo is misreading how Akamai works because he’s takes the word “network” too literally. Content delivery networks are not “networks” in the strict sense of physical infrastructure for moving data around. The Akamai “network” is just a bunch of servers sprinkled around the Internet. They use vanilla Internet connections to communicate with each other and the rest of the Internet. Internet routers route Akamai packets exactly the same way they route any other packets.

The “intelligence at the core of the network” Yoo discusses doesn’t actually exist in routers (which would violate network neutrality), but in Akamai’s magical DNS servers. DNS is the protocol that translates a domain name like techliberation.com to an IP address like 72.32.122.135. When you query an Akamai DNS server, it calculates which of its thousands of caching servers is likely to provide the best performance for your particular request (based on your location, the load on various servers, congestion, and other factors) and returns its IP address. Now, from the perspective of the routers that make up “the core of the network,” DNS is just another application, like the web or email. Nothing a DNS server does can violate network neutrality, just as nothing a web server does can violate network neutrality, because both operate entirely at the application layer.

So in a strict technical sense, Akamai is entirely consistent with network neutrality. It’s an ordinary Internet application that works just fine on a vanilla Internet connection. Now, it is true that one of the way Akamai enhances performance is by placing some of its caching servers inside the networks of broadband providers. This improves performance by moving the servers closer to the end user, and it saves broadband providers money by minimizing the amount of traffic that traverses their backbones. This might be a violation of some extremely broad version of network neutrality, and there’s certainly reason to worry that an overzealous future FCC might start trying to regulate the relationship between ISPs and Akamai. But Akamai is not, as Yoo would have it, evidence that the Internet is already non-neutral.

SHARE: 6 comments

I just finished a second read-through of Chris Yoo’s excellent paper, “Network Neutrality and the Economics of Congestion.” It’s an excellent paper, and in this post I’m going to highlight some of the points I found most compelling. In a follow-up post I’ll offer a few criticisms of parts of the paper I didn’t find persuasive.

One point Yoo makes very well is that large companies’ ability to coerce their media environment is often overrated. He points out that people often overestimate the ability of media companies to dominate the online discussion. He points out, for example, that fears that the AOL Time Warner merger would become an unstoppable online juggernaut turned out to be overblown. The merged firm turned out to have little ability to shape the browsing habits of AOL customers, and AOL continued to bleed customers.

Similarly, Yoo makes the important point that when evaluating the ability of a broadband provider to coerce a website operator, it is the broadband company’s national market share, not its local market share, that matters:

application and content providers care about the total number of users they can reach. So long as their total potential customer base is sufficiently large, it does not really matter whether they are able to reach users in any particular city. This point is well illustrated by a series of recent decisions regarding the market for cable television programming. As the FCC and the D.C. Circuit recognized, a television programmer ’s viability does not depend on its ability to reach viewers in any particular localities, but rather on the total number of viewers it is able to reach nationwide. So long as a cable network can reach a sufficient number of viewers to ensure viability, the fact that a particular network owner may refuse carriage in any particular locality is of no consequence. The FCC has similarly rejected the notion that the local market power enjoyed by early cellular telephone providers posed any threat to the cellular telephone equipment market, since any one cellular provider represented a tiny fraction of the national equipment market. Simply put, it is national reach, not local reach, that matters. This in turn implies that the relevant geographic market is a national one, not a local one. What matters is not the percentage of broadband subscribers that any particular provider controls in any geographic area, but rather the percentage of a nationwide pool of subscribers that that provider
controls.

Once the relevant market is properly defined in this manner, it becomes clear that the broadband market is too unconcentrated for vertical integration to pose a threat to competition. The standard measure of market concentration is the Hershman-Hirfindahl Index (HHI), which is calculated by summing the squares the market shares of each individual firm. The guidelines employed by the Justice Department and the Federal Trade Commission establish 1800 as the HHI threshold for determining when vertical integration would be a cause for anticompetitive concern. The FCC has applied an HHI threshold of 2600 in its recent review of mergers in the wireless industry. The concentration levels for the broadband industry as of September 2005 yields an HHI of only 1110, well below the thresholds identified above. The imminent arrival of 3G, WiFi, WiMax, BPL, and other new broadband technologies promises to deconcentrate this market still further in the near future.

To put this in concrete terms, if Verizon wants to twist Google’s arm into paying fees for the privilege of Google’s customers, Verizon’s relatively limited national market share (about 9 percent) doesn’t give it a whole lot of leverage. It is, of course, important to Google to be able to reach Verizon’s customers, but Google has enough opportunities in the other 91 percent of the broadband market that it wouldn’t be catastrophic if Verizon blocked its customers from accessing Google sites. Conversely, Google would have a strong incentive not to accede to Verizon’s demands because if it did so it would immediately face demands from the other 91 percent of the broadband market for similar payoffs. Which mens that Verizon, knowing that blocking Google would be a PR disaster for itself and that Google would be unlikely to capitulate quickly, isn’t likely to try such a stunt.

The analysis would be different if we had a single firm that controlled a significant chunk of the broadband market—say 50 percent. But there aren’t any firms like that. The largest appear to be Comcast and AT&T, with about 20 percent each. That’s a small enough market share that they’re unlikely to have too much leverage over web providers, which makes me think it unlikely that broadband providers would have the ability to unilaterally impose a “server pays” regime on them.

SHARE: 0 comments