Keynote Address: Netflix: Streaming Entertainment to 200 Million Members Around the World
summary
Netflix runs Open Connect OCAs that stream to ~200 million members and peak above 100 Tbps; caches run FreeBSD on commodity hardware and aim for dollars-per-gigabit efficiency. Key specifics: pre-positioned content and disk placement that gives ~15% throughput boost, flash + NVMe and dual-actuator HDDs, UFS for zero-copy sendfile, kernel TLS and NIC inline TLS encryption, TSO and mbuf coalescing to cut memory bandwidth, LRO with RSS sorting, per-CPU page caches, and NUMA pinning of NIC-local connection state.
AI-written summary · about
chapters (13)
- 0:00Start
- 10:26Storage Technologies
- 10:52Three Different OCA Flavors for Different Workloads
- 13:16Filesystem Technologies
- 14:52Some interesting context on data...
- 17:52Position on Disk
- 18:45We try to be smart about disk 1/0
- 22:03Driving efficiency
- 24:14Reducing memory bandwidth
- 33:52Hardware offloads
- 36:15PCle Optimizations: Coalescing small memory segments
- 37:36Enabling new platforms/designs
- 40:53Typical Netflix OCA Workload
related talks
45mNetflix's success: Combining collaboration, hardware monitoring & AI
42mHow Netflix Built a Real-Time Distributed Graph for Internet Scale
31mMastering Near-Real-Time Telemetry and Big Data: Invaluable Superpowers for Ordinary SREs
46mA Day in the Life of a Netflix Performance Architect
18mKeynote: Mind the Gap: Bridging Cloud Native Innovation with Real-World Use
10mLightning Talk: Envoy @ Netflix
39mSuper Fast Real-time Data Processing on Cloud-Native Architecture
34mLightning Talks