Tokyo Direct Hit: Netflix Team Explains the Technical Background Behind Live Streaming

2025-10-24 08:34:04

Abstract generation in progress

Imagine this, the grandest event night in history is about to arrive, with 65 million fans eagerly awaiting, heart rates accelerating. How can all viewers ensure they catch the live match on time every time? Netflix users are accustomed to on-demand viewing, where everyone can choose their preferred viewing time. However, with live events, millions of fans are eager to go online to watch in real-time. How does Netflix ensure that its members do not miss any exciting moments? When live traffic breaks streaming media records, Netflix's infrastructure faces the ultimate stress test. The author attended a presentation by the technical team at Netflix's Tokyo office, explaining the technological background behind Netflix Live streaming, and here are the key highlights of the report.

What is the difference between Netflix Live and VOD?

Live and Random Viewing (Video on Demand, abbreviated as VOD), is different in that members hope to watch live events instantly, meaning there is only a short time to recommend suitable live programs. Recommending too early will diminish the excitement, and appearing too late will miss the wonderful moments. Every second counts.

Carefully planned instant recommendations

To capture the excitement in real time, Netflix has enhanced its recommendation system to provide instant suggestions, offering members richer and more engaging signals to click play at the most critical moments. What is the challenge? How to send dynamic, timely updates to over a hundred million devices worldwide, where the crowd effect can overload cloud services. Simple linear scaling is neither efficient nor reliable. For popular events, it can also divert resources from other critical services. Netflix needs a smarter, scalable solution, rather than just adding more resources, so Netflix built its own system.

Due to millions of devices being online, the schedule for live events may change at any time. To keep everyone perfectly synchronized, Netflix has established a system that dynamically updates recommended content as the events progress.

Netflix found that it needs to balance three constraints:

Time: The duration required for coordinating updates. Optimized throughput: The capacity handled by cloud services. Calculation base: Various requests needed to provide unique updates.

By dividing the instant recommendations into two phases to solve the optimization problem, necessary data is pre-fetched, and the load is distributed over a longer period to avoid traffic peaks. When the live event starts or ends, a low cardinality message is broadcasted to all connected devices.

Simulation of the competition day scene

In addition to establishing new technologies that support real-time recommendations, Netflix has also evaluated existing systems, identifying potential traffic hotspots. By predicting high traffic during live events, they generated simulated traffic for game days to study how online services respond to these surges.

To ensure stable live streaming, the Netflix team simulated a “game day” scenario to test the server's performance under extreme load. The results showed that even when the core system performed excellently, unexpected traffic spikes could still occur before and after the live stream, such as simultaneous cache expirations or large-scale recalculations. To address this issue, Netflix introduced a “cache expiration jitter mechanism” that randomizes cache update times to avoid traffic congestion caused by centralized refreshes. At the same time, Netflix also established an adaptive traffic prioritization system that can automatically direct critical traffic to more scalable clusters during peak times and temporarily lower the priority of non-critical updates to ensure stable live streaming.

When asked during the live Q&A how Netflix could respond if Amazon Web Services experienced downtime, the team replied that Netflix has centers in each region that can handle distributed processing. The innovation behind Netflix is the collaboration of various cross-departmental teams, from engineering, cloud architecture, data science, search recommendations, to content delivery and device playback, each segment is put to the test. Netflix has extended from online streaming of movies and series to live broadcasts and gaming, and the technical team behind it states that they are continuously updating the technological level to enhance user experience. The future goal is to enable viewers not only to watch on demand at any time but also to participate in real-time interactive experiences, enjoying a seamless and immersive entertainment experience.