Design Live Video Streaming Service in 45 Minutes: System Design Interview Guide

Step-by-step system design walkthrough for a live streaming platform like Twitch/YouTube Live. Covers ingest (RTMP), transcoding, HLS delivery, chat, CDN, scaling, and reliability patterns.

Here is the detailed system design for a Live Video Streaming Platform (like Facebook Live, YouTube Live, or Twitch).

1. Problem definition and scope

We are designing a large-scale live streaming platform where users can broadcast video in real-time to a global audience. The system needs to handle video ingestion from content creators, process that video into different formats, and deliver it efficiently to millions of viewers.

• Main User Groups:

  • Broadcasters: Users streaming video (via mobile or desktop software like OBS).

  • Viewers: Users watching the live stream and interacting via chat.

• Scope:

  • We will focus on the Live Video pipeline: Ingestion (RTMP), Transcoding, and Delivery (HLS/DASH).

  • We will include basic Live Chat and Real-time Viewer Counts.

• Out of Scope:

  • Video on Demand (VOD) playback after the live event ends.

  • Monetization (Ads, Super Chats).

  • Complex recommendation algorithms.

2. Clarify functional requirements

Must Have:

• Broadcasting: Users can start a live stream using a unique stream key.

• Playback: Viewers can watch streams with acceptable latency (10–30 seconds) and minimal buffering.

• Adaptive Bitrate (ABR): The system automatically serves different qualities (e.g., 360p, 720p, 1080p) based on the viewer’s internet speed.

• Recording: The stream is automatically saved for later viewing once it ends.

• Chat: Viewers can send and receive messages in real-time.

Nice to Have:

• Viewer Count: Real-time counter of concurrent viewers.

• Reaction Overlays: Floating hearts or likes on the video.

3. Clarify non-functional requirements

• Scale:

  • DAU: 50 million.

  • Peak Concurrent Viewers: 5 million total.

  • Peak Concurrent Broadcasters: 50,000.

  • Single Stream Peak: Up to 1 million viewers on a single viral stream.

• Latency:

  • Targeting 10–30 seconds (Standard HLS).

  • We are not targeting sub-second latency (like Zoom/WebRTC) as it is prohibitively expensive for millions of viewers.

• Availability: 99.99%. Streams should not drop during broadcast.

• Read/Write Ratio: Extremely read-heavy. (1 broadcaster writes : 100,000 viewers read).