How Netflix Live Streams to 100 Million Devices in 60 Seconds

2025-12-15 · Source: ByteByteGo Newsletter · Field: Technology & Digital — Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Advanced, long

Summary

Netflix's Live Origin is a custom-built, multi-tenant microservice on AWS EC2 designed to manage real-time video delivery for live streaming at massive scale, handling events like the Tyson vs. Paul fight with 65 million concurrent streams. It employs redundant regional pipelines and intelligent segment selection to ensure quality, leveraging a manifest design with fixed "2-second" segment durations and millisecond-grain caching for Open Connect optimization. The system evolved from AWS S3 to a custom Key-Value Storage Abstraction built on Apache Cassandra and EVCache, providing high write availability, low-latency replication, and robust read scalability up to 200 gigabits per second. Key architectural decisions include publishing isolation, priority-based rate limiting for critical traffic, and hierarchical metadata caching to effectively mitigate "404 storms" and ensure reliable content delivery. This sophisticated architecture balances write reliability, read scalability, and operational flexibility for global live events.

Key takeaway

Netflix's Live Origin architecture, supporting 65 million concurrent streams, provides a blueprint for high-scale, low-latency live content delivery. It employs redundant pipelines, intelligent segment selection, and a custom Cassandra/EVCache storage system achieving 25ms median latency and 200 Gbps+ read throughput. This design offers critical insights for AI/ML professionals building resilient, real-time inference or data processing pipelines under extreme load.

Topics

• Netflix Live Origin
• Live Video Delivery
• Content Delivery Networks
• Apache Cassandra
• Distributed Caching

Best for: Software Engineer, DevOps Engineer

Related on AIssential

• Netflix Serves 84% of Query Results from Cache with Interval-Aware Caching in Apache Druid — Decomposing query results into time-aligned segments enables high cache reuse for rolling window analytics.
• QCon London 2026: From DVDs to Global Streaming How Netflix’s Commerce Architecture Actually Evolve — At QCon London 2026, Netflix Principal Engineer Kasia Trapszo detailed the evolution of Netflix's commerce architecture from a US-based DVD rental service to a global streaming platform across 130+ countries.
• Oracle Database at Azure: The Complete 2026 Technical Deep Dive Into Sub 200 Microsecond Multi… — Oracle Database at Azure provides low-latency, high-availability Oracle databases integrated with Azure's AI and data services.
• Embed a live AI browser agent in your React app with Amazon Bedrock AgentCore — Embed real-time AI agent browsing sessions into React apps for enhanced user trust and auditability.
• Building the AI Grid with NVIDIA: Orchestrating Intelligence Everywhere — AI-native services are exposing a new bottleneck in AI infrastructure, shifting the challenge from peak training throughput to delivering deterministic inference at scale with predictable latency and…
• Improve operational visibility for inference workloads on Amazon Bedrock with new CloudWatch metrics for TTFT and Estimated Quota Consumption — New Amazon Bedrock metrics provide server-side visibility into streaming latency and actual quota consumption.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ByteByteGo Newsletter.