Measuring the live video latency at scale
summary
A latency measurement framework for Linkedin Live measures per-stream latency nearline at scale and reports four components: ingestion-to-storage, origin processing, origin-to-CDN, and CDN-to-client. It embeds timestamps in fragmented MP4 via the PRFT box, pushes ingest events through an Event Pipeline, and fetches segments with a cloud Cloud Watcher using Azure Media Service integration. The system runs at scale, exposes clock-vs-media timestamp offsets, and surfaces real glitches found while building the flow.
AI-written summary · about
chapters (5)
related talks
19mLow Latency Live from a Different Vantage Point
44mLive Video in a Modern Web Environment
16mCDN Challenges of HTTP-based Low Latency Live Streaming Delivery
35mLive Low Latency Video Streaming
24mMeasuring the Reliability of Live Video Infrastructure
17mLive Video Streaming as "tail -f"
23mVideo @Scale Fall 2022 LIVE Q&A #2
19mTesting things that don't want to be tested: A game streaming story