What is Netflix paying for?

30 Apr 2014

Rob Sterling objects to this graph from Know more:

It shows the change in Netflix streaming speeds for various ISPs since January 2013:

Know more/Netflix graph

After a little back and forth on Twitter, I think I mostly understand what he means. Mr. Sterling thinks the chart shows that the growth of bandwidth hungry video streaming has outstripped capacity on the last mile. If that were the case, or if were the case for the ISPs’ regional backbones that connect different local areas to each other, then that would obviously be a problem that requires expensive upgrades, and someone would have to pay for that.

However, that doesn’t explain the sudden jump in the streaming speed of Comcast subscribers starting in February. Comcast did not perform any last mile upgrades. We are told that Netflix does not receive any prioritization on the last mile at all, so streaming growing faster than last mile capacity cannot explain why Netflix speeds improved.

What Netflix received was access to Comcast’s network at various peering points throughout the country. Switch and router ports aren’t free, so it probably cost Comcast something, but the cost was negligible compared to the fee they receive from Netflix or what actual last mile or backbone upgrades would have cost.

You may think that Comcast should charge Netflix in some way (I would disagree)—that even if Comcast doesn’t need the money for any immediate upgrades, the rents will be an incentive for some new entrant to displace Comcast. The only leverage Comcast has is to restrict peering access to Netflix or Netflix’s transit provider. But economically, the rents that Comcast would gather should not be seen as direct compensation for the capacity that Netflix is using. You could see it as simply adding to the rents that Comcast’s subscribers are paying, and if you love rents, you’ll love this one.

The only relevant capacity that could be outstripped is in the last mile, and that has not occurred.

When I pay Time Warner Cable for my 50 Mbps, I expect to get that at least speed to some TWC switch in the New York area. I know that it’s not guaranteed all the time all the way to any arbitrary host on the Internet. In my particular case, it’s a cable connection, so it’s shared with a bunch of my neighbors; nonetheless, we have collectively paid for whatever shared bandwidth the DOCSIS 3.0 equipment provides for, all the way to the TWC’s backbone, where Netflix or its transit provider would connect to under a peering arrangement.

Going back to the original chart, does it show that streaming grew faster than last-mile capacity? No. It shows that streaming grew faster than capacity at the ISPs’ peering points with Netflix’s transit provider. That’s why Cablevision and Cox are so much faster than the other ISPs—they don’t have a faster last mile, but they have settlement free peering and Netflix caches in their network. (Google Fiber peers with Netflix too, but they also have have a newer and faster last mile infrastructure.) That’s also why Comcast was able to quickly improve Netflix speeds without prioritization or changing anything at all in the last mile: they expanded peering with Netflix after being paid to do so, which is a very fast process.