The Comcast-Netflix Paid Peering Debate

Ask Dr. Peering

DrPeering -

What do you make of the Netflix paid peering relationship with Comcast? Is this good or bad for peering?

Thurston Howell III

Thanks for the question.

Much like a living organism, the Internet peering ecosystem responds and reacts to the stimuli placed upon it. One could view the two players on the main stage (Netflix and Comcast) as acting out the natural market response (Paid Peering) to the stimuli (massive video and all-you-can-eat charging model) placed upon it.

We see this dynamic manifested as two sides to a story, with both sides making seemingly valid points. If there was a great debate on this, I would expect at least the following arguments to be presented.

Netflix

----------

When I worked with the motion picture industry a few years back they shared that:

“Internet video is unlike other Internet applications in that even short-duration glitches (packet loss, jitter, etc.) lead to “artifacts” such as freeze frames, pixelation, garbled audio, etc. that destroys the “suspension of disbelief” that makes the movie experience enjoyable.”

I personally find it amusing that one of the first words my 3-year old learned was “buffering.”

Netflix wants the best possible end-user experience for its customers, an experience with no artifacts, and there are several interconnection techniques to do this.

The commodity Internet Transit model mostly works, but the more network components that there are in the path between content and eyeballs, the greater the chance of network glitches. Even if everything is perfect 99.9% of the time, if there happens to be a denial-of-service attack across the network path your video is traveling, you may experience video artifacts including the dreaded buffering indicator.

The interconnection techniques that minimize the chance of packet loss and buffering involve some blend of

1. Content Delivery Network (CDN) services,
2. Internet Transit (preferably from ISPs with eyeballs),
3. Internet Peering with eyeball networks where possible,
4. Paid Peering with eyeball networks, and
5. Remote / Distributed peering.

Each of these distribution methods has pro’s and con’s, but ultimately most content companies will employ some blend of these to offload the content directly to the eyeball networks.

Opening Statement

Historically, content companies have argued that the access networks should not charge them paid peering fees because the access network customers were the ones requesting the content!

The content companies simply want to effectively deliver that requested content in a way that maximizes the end-user experience. This seems like something that both Netflix and Comcast should simultaneously value, right?

Comcast

------------

On the other side, Comcast would argue that there is no free ride on the Internet. Everyone pays to get their bits across the Internet, and rightly so - there is a cost to operating networks.

Comcast must ensure the continued viability of its broadband Internet services. This means that the revenue model scales as its infrastructure costs grow. When Netflix or anyone “dumps” massive volumes of traffic onto the Comcast network, directly or indirectly, there needs to be a way for that traffic load to also generate revenue to cover the cost for the current and next major upgrade to the Comcast infrastructure.

In other words,

For sustainability there needs to be a way for the dollars to flow towards the bottlenecks.

The problem is that access networks bear the brunt of the costs of these video guys “business models” but the access networks don’t receive a dime for carrying this traffic. This is not sustainable. Paid peering is a way for Comcast to offset the cost of handling this massive video traffic distribution.

Wouldn’t it be great if you could get rich offering a service that exploits other peoples resources for free? It may be good while it lasts, but ultimately, no business will stay in business long doing this over the long term.

The Debate

-----------------

Content Providers would most likely counter that

1)the only reason that content providers are “dumping” that traffic onto the Comcast network is that Comcast customers are requesting it.
2)Comcast does get paid for that traffic, in fact, Comcast gets paid a lot by global standards for access to the global Internet, and the global Internet includes Netflix traffic.
3)Unlike the previous donut peering trend, there is no alternative way to reach Comcast customers but through Comcast’s network. There is no peering bypass possible here since most residents are singly homed to either DSL or Cable. So Comcast is exploiting its market dominance in this paid peering maneuver.
4)Everyone that uses Netflix has seen that “buffering” notification, so there are clearly performance problem delivering video from Netflix to Comcast customers.
5)The market for Internet service in the U.S. is clearly broken and it is time again to plea for repair.

Comcast would counter that

1)The current last mile infrastructure is under much strain today, mostly because it was designed before Netflix was 40% of all Internet traffic, before there were 3-6 video devices per home all streaming video at the same time. The all-you-can-eat business model with its current price points is simply not capable of handling everybody’s video traffic with the level of reliability required for flawless video.
2)Customers will switch to DSL if cable Internet prices are raised or if Comcast phases out the all-you-can-eat model. As a result, Comcast employs a two-sided model like the phone system - residents pay a little, and businesses like Netflix pay more.
3)The answer may the end of the all-you-can-eat model. The market is pretty good at handling these commercial disputes.

Content Providers then counter :

“So Comcast has a broken all-you-can-eat business model, and

the Comcast solution is to fix its broken business model by forcing content companies to pay to deliver content to Comcast’s captive customers.”

And so it goes, back and forth. Both sides have good points.

The Comcast-Netflix Paid Peering Agreement

If I was to guess, I would say that the paid peering agreement in question is probably a multi-year agreement that specifies a dozen or so specific points of n*40G interconnections, with a metered rate of $0.30/Mbps stepped down by maybe 3% per year. Additional locations will be added over time. The agreement would be protected under NDA of course, so neither side is allowed to talk about the agreement. OpenConnect is not mentioned in the agreement since Comcast doesn’t really want anyone else’s equipment in their network, and would not want to set the precedent.

The document describes procedures and timelines for incremental capacity over time. Both sides require quarterly coordination meetings to review performance statistics and to scheduled upgrades in advance of needing the capacity. Contact information and escalation procedures would be exchanged, along with instrumentation on either side of the interconnect. The other usual peering clauses are probably there as well. That is my educated guess anyway.

History Repeats Itself

So where does that leave the peering world?

Of course we saw that when Akamai caved in and started paying Comcast for paid peering, LimeLight reluctantly followed suit in order to compete with Akamai. Limelight couldn’t claim to get its customer’s content as close to the eyeballs as possible if they didn’t get right on Comcast’s network like Akamai did. Over time, the other CDNs fell in line too and all started paying Comcast. Like a stack of dominoes, once the first one fell, the rest followed.

This is Access Power Peering at its finest.

Looking forward, the Netflix competitors will most likely follow suit and purchase Comcast Paid Peering to provide the best possible experience for their end-users. And the other content companies will follow suit, or use services that have already followed suit. The ecosystem will have morphed again, demonstrating the magnificence and shifting shapes of this ecosystem organism.

This is my prediction anyway.