In 2008, Comcast announced a new policy to institute data caps on residential Internet use. Comcast subscribers who exceeded 250GB of data usage twice in a six-month span were cut off from service. While the policy was initially met with some criticism, the relatively generous cap did not affect the vast majority of subscribers, as the 250GB cap was, according to Comcast, enough to send some 50 million e-mails, download 62,500 songs, or download 125 standard-definition movies.

On May 17, 2012, Comcast announced that it was suspending the 250GB cap, allowing its subscribers unlimited use until the company implemented a new policy. Toward this end, Comcast initiated two test programs, one in Nashville Tennessee and one in Tucson, Arizona, to experiment with new data management practices.

In these cities, all users are allotted 300GB of data usage. Users who exceed the 300GB limit are charged $10 for an additional 50GB of data but will not need to worry about being cut off from service. Subscribers are also given three “courtesy passes” per year notifying them that they have exceeded their data allotment, but would not be charged the $10 fee.

Like the 2008 policy, the new program is unlikely to affect most subscribers, at least in the short term. According to Comcast, the median data usage is between 10 and 12GB per month. Streaming high-definition video uses approximately 2.3GB per hour, which means that you’d have to stream for about 130 hours before reaching the cap (assuming that the only thing you used Internet for was to stream video).

But here’s the rub: Everything you do on the Internet will count toward the data cap—everything except certain Comcast services. For instance, Comcast has stated that its Xfinity TV OnDemand service for Xbox will not count toward the cap. This means that users with the service could, in theory, stream 24/7 without ever running afoul of the data limitations. The same would not be true, however, for a Comcast subscriber streaming the same content using the Netflix Xbox app.

This has justifiably raised some eyebrows among net neutrality advocates who believe that the exclusion policy is inconsistent with the FCC’s Open Internet regulations. The policy has also raised antitrust concerns, prompting the Department of Justice to initiate an investigation into whether Comcast is meeting its obligations under the Consent Decree it entered into as a condition of its merger with NBC Universal.

Pursuant to the Consent Decree, Comcast “shall not prioritize [its] Video Programming or other content over other Persons’ Video Programming or other content.” Additionally, “[i]f Comcast offers consumers Internet Access Service under a package that includes caps … it shall not measure, count, or otherwise treat [its own] network traffic differently from unaffiliated network traffic.”
So how does Comcast square its policy with these provisions?

According to Comcast, the excluded content travels over Comcast’s own privately managed network infrastructure, which is “separate” and “distinct from” the “regular Internet.” Therefore, it doesn’t count against the data cap, which measures only data transferred over the public Internet. Comcast emphasizes that some of its own content, including that provided through XfinityTV.com and NBC.com, is counted against the cap, since, unlike the Xbox app, these services transmit data over the “regular Internet” rather than the separate lines used for Xfinity TV OnDemand.

Comcast’s position relies crucially on a distinction between the “regular Internet” and Comcast’s own “separate service flow[s].” However, Bryan Berg, founder and chief technical officer of Mixed Media Labs and a recognized technical infrastructure expert, has conducted a series of tests (see here, here, and here) that lead him to dispute Comcast’s claim of a “separate” Internet, and to argue that Comcast is favoring its content at the expense of other providers.

Responding to the claim that the Xfinity content is delivered via a “separate” Internet, Berg explains that “[i]f Comcast was using dedicated downstream channels to deliver Xfinity traffic,” as the company claims to be doing, then “only [certain] dedicated channels would increase when streaming video in the Xfinity Xbox app.”

However, this is not what Berg observed when streaming video on the Xfinity Xbox app. “Instead, like any other Internet traffic, the Xfinity traffic [was] directly observable to be balanced … across all … downstream channels…” Based on this, Berg concludes that the company’s claim of a “separate” Internet is “at best, stretching the truth.”

With regard to prioritization, Berg believes that “Comcast is using separate DOCSIS service flows to prioritize the traffic to the Xfinity Xbox app.” (DOCSIS is the technology that allows cable TV infrastructure to function as an Internet connection.) This practice allows Comcast “to exempt [Xfinity] traffic from both bandwidth cap accounting and download speed limits.” The manner by which the content travels is identical to that of other streaming services, he says, but “additional priority is granted to the Xfinity traffic at the DOCSIS level.”
To understand Berg’s point on prioritization, some background information is needed.

A standard and uncontroversial practice of network management is to artificially limit the amount of bandwidth available to a particular IP address. (An IP address is series of numbers that define a computer’s location on the Internet.) In contrast to the data cap described above, which limits the amount of data that can be transferred over the span of a month, this restriction limits the amount of data that can be transferred in one second. The extent of the bandwidth restriction can vary depending on the level of service one purchases from their Internet service provider. Berg’s plan, for instance, allows him a capacity of 25MB per second. Due to the artificial limit, Berg’s download speeds should never exceed 25MB per second even though his Internet service provider, which provides service to many others in the area, has sufficient capacity to permit much higher download speeds if left unrestricted.

In his tests Berg created local bandwidth congestion by starting twenty-four downloads at once. He then attempted to stream video using the Xfinity TV OnDemand and Netflix applications on his Xbox. While the Xfinity content was unaffected, the Netflix content required long buffer times and displayed low quality video. According to Berg, the difference is due to the Xfinity content not being subject to the artificial bandwidth restriction. Whereas the Netflix content was forced to compete for bandwidth within the 25MB per second limit, the Xfinity content was transmitted independently of the limit, reaching rates of up to 40MB per second. (See results for Netflix and Xfinity tests.)

While Berg’s analysis is compelling, it’s important to note that his observations do not address whether Xbox Xfinity content is similarly prioritized during periods of non-local network congestion—that is, during peak use hours when the combined use of all Comcast subscribers in a particular area approaches the provider’s maximum bandwidth capacity. Berg suspects that it is, but he admits that this is a question beyond the scope of his study.

Nevertheless, by Comcast’s own admission, its new policy excludes Xfinity on Xbox traffic from its monthly allotment calculations. And, if Berg’s observations are correct, Xfinity content delivered via the Xbox app is also unburdened by bandwidth restrictions that other content providers are subject to. Under these circumstances, the Department of Justice may indeed find that Comcast is “treat[ing] [its own] network traffic differently from unaffiliated network traffic” in violation of the Consent Decree.