Netflix And The Net Neutrality Promotional Vehicle

Federal Communications Commission chairman Tom Wheeler says he’s looking into Internet interconnection practices in general, and the disputes between Netflix and some broadband providers in particular. Wheeler even visited Silicon Valley Wednesday to discuss the matters with big technology firms.

Netflix is asking Washington, D.C., to step in and regulate the Net as never before. Exhibit A in Netflix’s case last winter was that the streaming speeds on some ISPs, according to its own speed index, had dropped. It accused the ISPs of deliberately slowing traffic. Before we decide to regulate our most vital economic platform, however, we might want to investigate. A network is only as fast as its slowest link, and the journey of a packet on the Internet is often long and winding.

Netflix's ISP Speed Index implies it is a measurement of the speed of the broadband service provider’s last-mile connections to residential customers. It is not. It is a measure of the average bit-rate of Netflix streams that customers of various broadband providers experience. But it may have relatively little to do with the last-mile broadband network itself.

For example, an Ookla speed-test with Verizon FiOS may show that the last-mile link is operating at the advertised 50 megabits per second. But Netflix’s ISP Speed Index may show a Netflix watcher using a Verizon FiOS broadband connection is streaming at 2.20 megabits per second.

Why?

First, a Netflix stream only requires a couple megabits per second — 2-3 megabits for standard definition (SD) video, perhaps 6-10 megabits for high definition (HD). So if everything works perfectly, Netflix’s ISP ranking may only show a “speed” of 2-4 megabits per second. This isn’t a slow connection. A perfectly good stream only requires that much bandwidth. Remember, Google Fiber, the top ranking broadband provider in Netflix’s ranking, has advertised capacity of a gigabit per second but registers “just” 3.41 megabits per second on the Netflix ranking. That’s less than four tenths of one percent of Google Fiber’s capacity.

Second, the choice of content will affect the ranking. If ISP A’s users (or their devices) tend to choose more standard definition content but ISP B’s users and devices tend to choose more high-def content, ISP B will appear faster in the ranking. Yet the difference may not have anything to do with the quality of the broadband connection.

Third, the quality of one’s home network also affects the experience. If I’m watching on a portable device in my backyard or an upstairs bedroom, my home Wi-Fi network may not have the reach to deliver optimal performance. The broadband line to my house may have plenty of capacity, but the last-fifty-foot wireless link to my device may not.

Fourth, the routes that Netflix uses to deliver its content to the broadband providers may be suboptimal. Netflix has used at least four network architectures over the years, and it still uses several of them today. It has used third party content delivery services (CDNs) like Akamai and Limelight; third party backbone networks in conjunction with its own CDN, OpenConnect; embedded content caches within some small broadband providers; and direct “on net” or “paid peering” connections with broadband providers. And these distinct paths don’t always perform equally well.

Sandvine showed that Netflix often chooses among these different paths depending on the device being used to watch. Using an iPad, for example, I might get a stream via one of Netflix’s retail CDN providers like Akamai. My daughter, who is using the family PC, on the other hand, might get a less reliable stream from a backbone network.

A NetForecast examination of the Netflix ranking data shows that Netflix’s agreement early this year to connect directly to Comcast improved performance not only for Comcast customers but also for 11 other ISPs who had a also suffered slow-downs. How so? The direct Netflix-Comcast connection removed around 30% of the traffic Netflix was trying to push through its too-small backbone network and thus relieved congestion on the links to those 11 other ISPs.

This shows that direct “on net” or “paid peering” connections, like the Comcast-Netflix hook up, improve performance. But it also shows that Netflix’s own stingy network architecture, in which it tried to push too much traffic through an indirect and undersized backbone path, was the key culprit in its content slow-downs.

Peter Sevcik of NetForecast concluded that “the Netflix index largely documents user and Netflix choices, not ISP performance.”

These details matter because Netflix (and a few other firms) are trying to get the Federal Communications Commission to change the way the Internet works — in both its commercial arrangements and architecture — to suit their own business models. But Washington, D.C., should not dismantle a a system that works remarkably well, and which has repeatedly demonstrated the ability to adapt, to favor one firm’s bottom line or to bail it out from poor network architecture decisions. The “strong net neutrality” regulations Netflix is seeking would make illegal many of the connections among networks that have existed for years, or decades.

In 2006, Netflix, without admitting wrongdoing or liability, settled a class action suit that accused the firm of “throttling” some customers — denying them the “unlimited” DVDs it promised, and instead favoring light users over heavy users when making inventory decisions. Netflix agreed to give current and former customers a free month of DVDs, but in its original form, the settlement would have, after the free month, automatically continued customers’ service at a new, higher rate. The Federal Trade Commission criticized the original settlement terms, saying it was less compensation than it was “a promotional vehicle.”

Promotional vehicle. That’s a pretty good description of this latest episode of self-induced throttling and policy grandstanding.