‘Preliminary’ MIT/UCSD Study Finds Internet Congestion Not Widespread

Network congestion on the Internet “does not appear to be widespread” and is mostly confined to “recognized business issues” like Netflix’s complaints of streaming latency, a “preliminary” study by the Massachusetts Institute of Technology (MIT) and the University of California, San Diego (UCSD) said Wednesday. Congestion that does exist “can come and go essentially overnight” due to provider-initiated network reconfigurations and changes in content routing, the study said (http://bit.ly/1oGQoJG). The results are “very early” findings in a project that will eventually become an “atlas” of congestion on the Internet, said David Clark, senior research scientist at MIT’s Computer Science and Artificial Intelligence Laboratory, at a Congressional Internet Caucus Advisory Committee event marking the study’s release.

Clark referenced specific congestion incidents included in the study, conducted using probes owned by UCSD’s Cooperative Association for Internet Data Analysis (CAIDA) at 10 access points. In one example, the study found periods between February 2013 and March 2014 when Cogent, Level 3 Communications and Tata all experienced spikes in congestion at interconnection points with Comcast in the San Francisco Bay area. All such congestion dropped off dramatically after Comcast and Netflix signed a peering agreement in February, the study said. Clark noted that the congestion indicated a correlation rather than a causation.

The study included measurements from four major U.S. ISPs -- Comcast, Cox Communications, Time Warner Cable and Verizon -- along with content providers like Netflix and YouTube, content delivery networks like Akamai and Limelight, and international ISPs like BT and Free. Most congestion measured in the study was of a short duration, and occasional congestion “hot spots” aren’t surprising given the need for constant network upgrades, Clark said. Given that measurements for the study come from one or two vantage points per ISP, it’s likely the study hasn’t hit all interconnection paths between the studied entities, he said.

Industry participants and observers at the event framed the study’s findings in the context of interconnection-related business disputes, including Netflix’s disputes with Comcast and other ISPs. Netflix Vice President-Content Delivery Ken Florance said he agreed that most ISPs do not have persistent congestion issue but added that some of the largest ISPs have been allowing congestion issues deliberately. Ike Elliott, CableLabs senior vice president-strategy, said he agrees with the study’s preliminary results, noting that it shows that much of the congestion occurs at interconnection “on-ramps” to the ISPs and that entities sending the traffic -- like Netflix -- are at fault if they choose the most congested on-ramps.

Public Knowledge President Gene Kimmelman said the congestion the study highlights “calls out” for policy oversight that bleeds over from antitrust concerns into the public interest given that the congestion ultimately affects the user experience. Those policy solutions should not ultimately lead to regulation of the peering market, said Jeff Eisenach, director of the American Enterprise Institute’s Center for Internet, Communications and Technology Policy.