Image source: Netflix.
Unless you've been living under a rock in the Mojave desert for the past few years, you already know that Netflix (NFLX -3.92%) cares a lot about high-speed networking. It sort of comes with the territory when you're running a digital video service that can consume more than one-third of all Internet bandwidth at peak viewing hours.
This big appetite for data bandwidth also gets Netflix in hot water with Internet service providers like Comcast (CMCSA -0.37%) and Verizon (VZ -0.53%). To work around these conflicts, Netflix used to rely on a bevy of specialized content delivery networks owned by Level 3 Communications (LVLT), Akamai (AKAM 0.42%), and Limelight Networks (EGIO -9.23%).
But that wasn't good enough. Now, Netflix prefers to install its own CDN solution, known as Open Connect, directly into major networking hubs. Service providers don't always like that idea, even if the Netflix server is sitting just a couple of hardware racks away from the Akamai and Limelight solutions.
What's the big idea behind Netflix's rolling its own content delivery solutions? And if Open Connect really is all that and a bag of chips, why isn't every media publisher doing exactly the same thing?
With these questions burning at the back of my brain, I called up digital-media expert Dan Rayburn at consulting firm Frost & Sullivan. As a leading authority on media technology issues, Rayburn delivered exactly the kind of fresh insights I was craving.
Digital media expert Dan Rayburn. Image source: Dan Rayburn.
Why should Netflix run its own delivery network?
"There are lots of things that makes Open Connect different," Rayburn told me. "What Netflix has is what we in the industry call a purpose-built CDN. And that simply means it's built for one purpose: Netflix's content.
"You can't compare it to CDNs like Limelight and Akamai, because those guys have to support content from a lot of different providers that use different protocols, they have to support streaming, downloads, small object delivery, large object delivery, security applications, commerce stuff, and so on.
"Netflix only has to support one thing: on-demand video. So, it's much easier to scale Netflix's network and to build it out.
"However, at the same time, you're also talking about a network that has to have a lot of capacity and a lot of reach, because it takes up such a huge amount of peak Internet traffic at night."
In other words, it's a matter of scale. Yes, the CDN specialists do offer specialized delivery services for video content. But even these titans of content delivery were never really built for the torrential traffic that Netflix videos create in Prime Time viewing hours, so they don't really have a suitable solution for Netflix's specific needs.
Doing one thing brilliantly is simply easier (and often cheaper) than doing many closely related things very well. And at Netflix-sized scale, "very good" just isn't good enough. So the company needed a better networking infrastructure, and the best way to achieve that was to simply create exactly the right thing.
Image source: Netflix.
What's the special sauce?
"Netflix Open Connect is similar to the other CDNs in that the fundamental building blocks of how this stuff works is all the same," Rayburn said. "You've got storage, you've got origin storage, you've got caches. You gotta have transit, interconnects, cross-connects, you're sticking your servers in a co-location facility.
"But Netflix's stuff is basically just optimized for them, so they're never all of a sudden going to get into the business of renting out what they've built and sell it to other content owners like Limelight does.
"So is there a reason to use Open Connect instead of a video CDN product? Yeah, you can't use Netflix's CDN unless you're Netflix. You can't use Apple's (AAPL 1.27%), or Microsoft 's (MSFT 0.37%), or Twitch. They're all purpose-built."
The basic technologies at the heart of Open Connect are very similar to what you'd find inside other CDN providers. But the system will only do exactly what it was designed for, which is to deliver the kind of digital video media that Netflix serves up, and in exactly the technical formats that the company prefers.
There's no dead weight here, and every piece of the system was explicitly designed to fit into Netflix's media management process. New content appears when Netflix lands a license for it, and falls away when that license expires. It's automated to the hilt, with multiple layers of error handling to ensure smooth operations at all times.
You simply can't expect that kind of custom fit from Level 3 or Akamai. What Netflix needs, only Netflix itself can deliver. The company's unique size and bandwidth heft makes any other solution impossible in the long run.
Where are all the copycats, then?
Netflix is fairly open about its Open Connect solutions. The hardware designs are published for anyone to see. Moreover, the company shares much of its custom software under an open-source license. In other words, anyone can essentially build a carbon copy of the Open Connect system at the drop of a hat.
So why isn't that happening today? Dan Rayburn had an answer for that stumper, too.
"Comcast has done the same thing as well," he said. "A couple of years ago at StreamingMedia.com's Content Delivery Summit, Comcast started doing presentations about the open source network that they built internally, and a lot of the tools that they used are open source. So they published a lot of that as well. Netflix definitely isn't the only one.
"The issue is, there's just not a ton of content owners out there who can really benefit from taking what Netflix has and trying to replicate it because from a CapEx and OpEx standpoint, it doesn't make sense to build your own CDN unless you're Netflix, Apple, Microsoft, Google, Twitch, Comcast...
"You know, Major League Baseball isn't going to build its own CDN because they can't do it as efficiently or as cheaply as the CDNs that they use."
The cost issue makes sense. There's no need for a small-time video wrangler to build a massive, custom content delivery system. In most cases, off-the-shelf services from Level 3 or Limelight will do just fine.
I was caught off guard by Rayburn's reference to Major League Baseball, though. The sports league offers unique video services and caters to a huge fan base. Wouldn't a media service on that level be able to use an in-house content delivery model?
Size plus expertise equals opportunity
Rayburn explained that custom networking lies too far outside MLB's circle of competence.
"Keep in mind, Netflix has a lot of in-house technical resources, and has for a long time," he said. "They've been well versed with very extreme technical people in house, because think on the whole DVD side of how they had to get that automated system down. They've really been a technology company from the start.
"So it wasn't shocking at all that they decided that they were going to build their own CDN, because they have that skill. They have a very deep pool of serious engineering talent over at Netflix.
"As do other companies, by the way. Someone like Major League Baseball Advanced Media have been doing this forever. But even they don't have their own CDN.
"And when you think about, say, Hulu -- that's not Hulu's specialty. That's not Hulu's business. That's not what they're good at. What Hulu is good at is licensing content, packaging it, making a good player, and marketing that service. Their specialty isn't building out infrastructure networks.
"So that's the other thing that companies look at. Do we have the expertise in-house to do that? Is that really our core competency?
"And for most of them, the answer is no.
"Hulu simply doesn't deliver enough traffic. If you look at what Hulu is at the peak hours of traffic, it's not a lot. It's a very small percentage."
Indeed, it is. According to the latest digital video report from Sandvine, Netflix streams uses 35% of all downstream traffic to North American broadband households at prime viewing hours. YouTube sports about half of Netflix's bandwidth hunger, and nobody else comes close.
Hulu almost looks like a rounding error with just a 1.4% share of primetime traffic. Major League Baseball wasn't even mentioned by name in that Sandvine study.
So most online media services can get by just fine with standard CDN solutions like Akamai or Level 3. Apple runs its own service to keep its App Store and iTunes media services humming along, and Microsoft runs a similar play for Windows updates and Xbox-related content.
When Hulu is too small to matter, despite more than 6 million subscribers and even more users of its ad-supported version, you know it would take an epic upstart to follow in the footsteps of Netflix and its Open Connect network.
