Medley: Optimizing Midgress Bandwidth for Commercial Live Streaming CDNs
summary
Medley implements a substream-based CDN that splits each live video into k substreams so each edge node caches 1/k of the stream, claiming a 76.83% midgress bandwidth reduction in production. A utility-driven K scaling component adjusts k in real time trading midgress gain vs. overhead; a hybrid scheduler assigns frames by type and size to avoid I-frame overload; and first-go isolation and fast request recover missing substreams. Deployed at ByteDance, it handles ~0.85M concurrent viewing requests per minute in live traffic.
AI-written summary · about
chapters (10)
- 0:00Medley and live streaming CDN
- 2:25Midgress traffic and costs
- 3:30Midgress waste in CDNs
- 6:39Medley’s substring splitting
- 8:50Implementing Medley challenges
- 10:14Dynamic substring scaling
- 11:18Balancing frame distribution
- 13:07Mitigating subscreen blocking
- 14:14Production results and savings
- 15:20Key takeaways and future
related talks
18mDelivering Reliable Live Streaming Over Unreliable Backbone Networks
18mMidgress-aware traffic provisioning for content delivery
24mPeople make the difference, or video streaming
14mAnchorNet: Bridging Live and Collaborative Streaming with a Unified Architecture
19mLive Media Over QUIC
51mHigh Bandwidth Data & Low-latency Java
21mStateless Datacenter Load-balancing with Beamer
19mDChannel: Accelerating Mobile Applications With Parallel High-bandwidth...
