How to Scale with WebRTC

See more videos from 流媒体 East Connect on the 流媒体 YouTube channel.

Learn more about WebRTC at 流媒体西部 2020.

Read the complete transcript of this clip:

克里斯·艾伦: The thing that people will usually tell you is that WebRTC doesn't scale. And that's absolutely true. If you look at it from a traditional standpoint, it was originally built as a peer-to-peer protocol, which inherently doesn't scale well. But what we do is we use cloud instances. As you can see in this diagram, we've got origin instances. On AWS there'd be EC2 instances. The incoming streams go into origins, they get fanned out to edges, and then subscribers are connecting to those edges. 然后, to get a massive concurrency, we put relays in the middle, which allows extra capacity one origin would be able to handle on its own. This is how you can get millions of concurrent viewers using cloud.

The brains of the operations is this thing called the Stream Manager, which is routing the streams. Broadcasters connect to it, and get assigned an origin. 订阅rs connect to it, and get assigned an edge to subscribe to that stream. And it takes care of auto-scaling. It can move up and down and add instances as needed depending on the load.

We also have a unique approach to ABR. Our system will actually re-react and send the correct data down based on this REMB message, which is happening over the WebRTC protocol. And probably more interesting is that we can actually leverage multiple cloud systems to get a really nice global coverage. So if one cloud provider is not gonna cover the regions you're looking to hit, then you can mix and match them. And we do this using Terraform, which is kind of this general API. It allows you to tap into just about any cloud network, which also includes private networks. So you can have your private cloud connecting to public cloud for excess capacity and everything else.