Release After a Year of Development
summary
A B2B real‑time video platform released a rushed December build and spent ~40 days chasing intermittent stream drops. Engineers used strace, perf, flamegraphs, network tests and VisualVM to narrow causes to DMA/NIC driver issues and overloaded origin servers, then replaced NICs, migrated hosts to CentOS7, added GPU transcoding (claimed ~200× cheaper) and during a seven‑day hackathon rewrote ingestion/transcode/control layers — resulting in a stabilized platform and a partner reporting ~7% higher conversions.
AI-written summary · about
chapters (10)
- 0:00Release overview
- 5:41First version issues
- 10:14Debugging first steps
- 16:36System calls and locks
- 23:56OS and network tuning
- 30:23Kernel and NIC tests
- 36:26Thread drops root cause
- 38:52Hackathon and infra rewrite
- 40:38Findings and advice
- 42:51Conclusion and Q&A
related talks
10mMoving Live Video Quality Control from the Broadcast Facility to the Living Room
18mHow we accelerated Live 2 VOD delivery from 15 minutes to under a minute
43mConference Recording and Streaming
16mSpontaneous Togetherness
20mThe NeverEnding Story of replacing your transcoding pipeline
30mTaking Ruby to the Movies
04mWhy build a new native player?
21mThe challenges of deploying Apple's Low Latency HLS In Real Life