Study Finds Internet Congestion Really Is About Business, Not Technology


Various enormous corporations have this year been at each other’s throats over how well or how poorly internet traffic travels through their systems. A new report indicates that some of the mud-slinging this year is true: interconnection, or peering, between ISPs is why end-users are getting terrible internet traffic. But, they say, it’s business, and not technology, that’s making your Netflix buffer.

DSL Reports points the way to the study, from an internet research organization called M-Lab. M-Lab studied how traffic does (or doesn’t) make it to you through the peering connections it travels through.

Peering has come up a lot this year, most notably around Netflix. The streaming-video behemoth contended that major ISPs — particularly but not solely Comcast and Verizon — were deliberately letting Netflix traffic clog up.

The congestion was happening at interconnection points, the places where the transit ISPs Netflix partnered with — companies like Level 3 and Cogent — met up with the access ISPs (the “last mile” providers) that home users use. To alleviate the congestion and avoid hemorrhaging customers who couldn’t use the service they paid for, Netflix eventually paid Comcast, Verizon, and Time Warner Cable for direct access to their networks, bypassing those interconnection points entirely.

But even though Netflix moves a lot of data, they’re hardly the only internet traffic moving around out there. Massive amounts of information pass through transit ISPs and peering connections every moment of every day. So where are the problems in the system?

M-Lab sat down and did a long-term study measuring how internet traffic moves through all those different transit tributaries, so to speak. The full report (a href=”http://www.measurementlab.net/static/observatory/M-Lab_Interconnection_Study_US.pdf”>PDF) delves into some of the technicalities of measuring and quantifying interconnection and is kind of a hefty read. There are, however, some clear key take-aways.

From a high level, you see some of the patterns you’d expect. For example, there’s a lot less network congestion at 3 o’clock in the morning then there is at prime-time, between 7 and 11 p.m.

But beyond that, one pattern began clearly to emerge. The network congestion M-Lab was seeing, they write, doesn’t appear to be connected to the technical limitations of ISPs or the connection points themselves. Instead, it seems, “business relationships between ISPs, and not major technical problems, are at the root of the problems we observed.”

M-Lab is very careful to remind readers that they have data proving correlations, but insufficient information to assign blame. “While we feel safe pinpointing the interconnection relationship … as a factor in performance degradation,” M-Lab writes, “it is important to not that we cannot determine which actors or actions are ‘responsible’ for observed degradation. We cannot tell whether any particular ISP between the user and a measurement point is ‘at fault,’ what the contractual agreement between ISPs did or did not dictate vis-a-vis interconnection, or whether specific network modification was done to alleviate or magnify a given incident.”

But the evidence does look worse for some companies than for others. M-Lab used New York City as one of their case studies. They found that in 2013, internet traffic delivered to Verizon, Time Warner Cable, and Comcast via Cogent took a dramatic hit before rebounding equally dramatically roughly 10 months later:

Since all three ISPs took a hit at roughly the same time, it seems easy to look for a common link. In this case, that link would be Cogent. Perhaps the trouble was on their end, instead of at three different companies?

Not so, says M-Lab. The next graph demonstrates that during the same period of time, in the same city, Cablevision’s traffic through Cogent moved smoothly and quickly without taking the same precipitous plunge as the other three providers’.

So that says Cogent was up and running. So perhaps TWC, Comcast, and Verizon all suffered a freak simultaneous set of outages? M-Lab couldn’t rule that out, but they also measured the connections all four last-mile ISPs had to another transit provider, Internap, during the same time period. Those connections did not see any significant degradation.

Although they are once again careful not to assign blame, M-Lab points out that the data “strongly suggests that the issues seen are not isolated to the consumer-facing, edge networks of impacted Access ISPs, but rather relate to the interconnections traversed between Cogent and the three impacted Access ISPs – Comcast, Time Warner Cable, and Verizon.”

Whether the last-mile providers are doing it to squeeze more money out of video companies or simply having contractual disagreements with Cogent doesn’t really matter to consumers who, once again, are stuck in the middle without the chance to flee to the competition.

M-Lab: Streaming Issues Were Conscious ISP Business Decision [DSL Reports]