What scales in a streaming service
- Application servers: sign-in, catalog, search and entitlement checks
- Playback authorization and DRM license issuing
- Payment and checkout
- Transcoding workers, which scale with the upload queue
- Live transcoding and packaging
- Video delivery itself, which is scaled by the CDN, not by your servers
How it works
The system watches indicators such as processor load, request rate or queue length. When one crosses a threshold, it starts more instances and adds them behind a load balancer; when demand subsides, it removes them. Databases and caches scale differently, through replicas and managed services sized with headroom.
Its limits: the live spike
Reactive scaling takes a few minutes to bring new capacity online. A live event can send most of its audience in under that. The answer is to scale ahead of time: raise capacity before a scheduled event, keep it there until the audience has arrived, and let it fall afterward.
Any component that does not scale, such as a single database or a third-party payment service with a rate limit, becomes the bottleneck. Load testing before large events finds these.
Why it matters commercially
Without it you either pay all month for capacity you need one evening a week, or you fail on that evening. With it, cost follows demand. On a managed platform, scaling is the vendor's responsibility and is part of what the plan buys.
Last reviewed