Hacker News Reader

Comment by cheald | Hacker News Reader

No, they didn't. Netflix had plenty of options for transit to Comcast - they didn't have to use Cogent. But it would have cost them a lot more. Cogent is famous for both cut-rate transit and peering disputes with last-mile providers.

The fact that they bought from Comcast rather than someone like Level3 or XO should tell you that they rate they got from Comcast was better than what they'd have gotten from another transit provider.

If you serve content on the internet, you have to buy transit from someone.

Edit: Would the downvoters like to point out what's wrong about my assertion?

View on HN

Netflix didn't buy transit from Comcast; they paid Comcast some large fee to upgrade links to one or more transit providers that Comcast would in the ordinary and traditional course of business done itself, so as to continue delivering its traditional unrestricted internet product to customers.

Netflix can't reach every Comcast node around the country, so it couldn't have 'bought transit' from Comcast. It did try the next best thing: colocating, at its own expense, servers at Comcast POPs with cached Netflix content. Comcast said no thanks to that one. Netflix also tried buying extra capacity from other transit providers (Tata and some others.) I think -- but am not certain -- that, when Comcast learned that that deal had happened, it began failing to upgrade those links, too.

It's hard to think of an explanation for these peering disputes between CDNs/transit providers and consumer ISPs that doesn't imply ISP hostage-taking. (Where the consumer is hostage.) These consumer ISPs have an easy job: keep these links to the middle networks and backbone networks up to date. And keep the copper in decent shape.

> they paid Comcast some large fee to upgrade links to one or more transit providers

Do you have a link for that? I was unaware of any of that happening. It was my understanding that Netflix is engaging in paid peering directly with Comcast now (http://bgp.he.net/AS2906#_peers).

You're right that Netflix isn't technically buying transit from Comcast, but they replaced their transit purchased from Cogent with paid peering purchased from Comcast; the net effect is the same.

> It's hard to think of an explanation for these peering disputes between CDNs/transit providers and consumer ISPs that doesn't imply ISP hostage-taking.

I think that's exactly what's happening here, but it's done as a bargaining chip in the never-ending peering wars, rather than Comcast specifically engaging in rent-seeking from Netflix. The history of settlement-free peering is long and sordid, and last-mile providers have been at odds with the T1s for decades. Netflix massively amplified that dispute simply by virtue of the amount of bandwidth they consume.

Netflix does use Level 3. At least for their connections to Verizon FiOS customers. Who also observe shit performance.

The Verizon/Level3 Netflix performance issue is also the consequence of a peering dispute between Level3 and Verizon, by admission of both Verizon and Level3 - not a consequence of Verizon specifically throttling Netflix traffic.

It's very important that the distinction be made between saturated interlinks and throttling of specific origins; many people seem to think that it's the latter that's been done here, when the issue is consistently interlink capacity disputes between last-mile ISPs and T1 providers.

To put it another way, unless these new rules require that ISPs hook up as much interlink as their T1 peers ask for, these rules won't fix shitty Netflix performance on last-mile networks who are engaged in peering disputes with Netflix's transit providers.

"To put it another way, unless these new rules require that ISPs hook up as much interlink as their T1 peers ask for, these rules won't fix shitty Netflix performance on last-mile networks who are engaged in peering disputes with Netflix's transit providers."

100% agree. But it isn't really "as much as peers ask for;" it's more like: as much as consumers demand, keeping peak link utilization at 50% or some industry standard figure.

Reply on news.ycombinator.com