Washington, DC -- Early data compiled by one of the country's premier technology universities indicates that the points where US Internet service providers link up their networks are generally free of abnormal congestion -- except when Netflix becomes part of the equation.
The preliminary report from the largest research laboratory at the Massachusetts Institute of Technology, released Wednesday in partnership with the US Congressional Internet Caucus, attempts to clarify the facts behind the complicated debate of Net neutrality, the concept that all Internet content should be delivered without preference or discrimination.
But the early data may only fuel ongoing bickering about a topic that is complicated by not only a lack of mainstream familiarity about how the Internet works but also the absence of transparency about the deals connecting networks.
Net neutrality has featured in sharp debates this year after a federal appeals court threw out regulations in January. The Federal Communications Commission then proposed replacement rules last month that some critics say could curb freedom of speech on the Internet. Many Internet companies oppose regulations that would allow ISPs to set up paid "fast lanes," saying that it would threaten innovation, if not the Internet itself.
Industry debate divides Netflix, ISPs
The debate has metastasized to include a high-profile conflict between Netflix and ISPs like Comcast and Verizon over interconnection agreements. These are deals between networks to connect their systems and more efficiently deliver content -- just the kind of links that the MIT study examined.
Netflix's video streams create more downstream peak Internet traffic than any other single company in North America. But the company has traded charges with Comcast and Verizon, two of the biggest US Internet providers, over who is responsible for degraded Netflix streams and whether Netflix should pay to interconnect. Both sides say the other is to blame. Netflix thinks it should be able to interconnect without paying, whereas Comcast and Verizon say the fees are nominal and this is simply how these deals have worked for years.
On Wednesday, Internet scientist and senior MIT researcher David Clark said that early data from his lab uncovered few signs of widespread congestion in the US. He said that most of the congestion was found at interconnection points presumably caused by existing business relationships.
In the last few months, the preliminary report traced most of the congestion to adjustments and negotiations related to the delivery of Netflix content into networks. Peering links -- connections where networks agree to swap traffic -- with Netflix appear to be congested for 18 hours a day. But the apparent congestion essentially vanishes when new interconnection links are put in place, likely due to business arrangements like the one between Netflix and Comcast, Clark said.
Clark warned that these generalizations should be taken cautiously: The data examined four major US access ISPs -- not all of them -- and did so over only a couple of measurement periods in January and in April.
Disregarding Netflix traffic, MIT's measurements found only occasional points of congestion where ISPs interconnect -- typically two or three congested links per ISP for about one or two hours a day. Congestion tended to be on costly links, like those that must travel under oceans.
A panel of representatives from Netflix, the cable industry, a conservative think tank, and a public advocacy group discussed implications of the data at the event here, which on the whole continued the same rhetoric that each entity has been trumpeting all along.
Ken Florance, Netflix vice president of content delivery, said he both agreed and disagreed with the generalization that congestion is not widespread. In most markets with most networks, Netflix has found no persistent congestion issues, but a few of the largest residential providers in the US "deliberately and systematically allowed their connections to congest," he said.
"A few top networks were purposely degrading their networks to get money out of content providers," he said.
Streaming video provider 'not the victim'
Ike Elliot -- a strategy executive from CableLabs, a nonprofit research-and-development firm paid for by a consortium of cable companies -- said the MIT data illustrated how Netflix has been the one manufacturing congestion by only choosing "on-ramps" to ISPs' networks that are backed up, a common cable industry claim in the interconnection squabble.
"The fundamental technology point is that the streaming video provider is not the victim," he said. "The streaming video provider is manufacturing or creating some congestion by choosing the traffic route."
In an unusual turn, Netflix said it was open to collaboration with the likes of Clark at the MIT Computer Science and Artificial Intelligence Laboratory, as well as another participant in the panel, the Cooperative Association for Internet Data Analysis based at the University of California at San Diego, to bring more transparency to their measurements of congestion.
"We tend to have a tremendous amount of data," Florance said. "I don't necessarily think that ISPs need to [share]. We're actively working with Caida, a number of organizations."
Typically, Netflix is fiercely protective of its data. It declines to share any viewership numbers for its popular original series like "House of Cards," for example -- not even with the show's creator.
The data and discussion came the same day as the FCC released its own figures on Internet service and congestion in the US. The FCC found back-ups at certain "peering" points in the network that likely affected consumers' quality of service.
"During our testing we noticed that at certain points through the test-network we saw some very serious congestion," an FCC official said during a conference call with reporters. "We aren't prepared to make conclusions right now, but we will be looking at how video service providers are affected by this congestion."