4.1 Design a Live Streaming Platform

This is a classic system design question that tests your ability to handle real-time data at massive scale. Live streaming sits at the intersection of storage, networking, and distributed systems — and interviewers love it because there are so many interesting trade-offs to discuss. **Estimated time: 45 minutes** --- ## Step 1: Understand the Requirements (5 minutes) Before you touch a single architecture box, you need to nail down what "live streaming" actually means for this system. Ask your interviewer clarifying questions — it shows maturity and prevents you from designing the wrong thing. ### Functional Requirements - **Broadcasters** can start a live stream and upload video in real time - **Viewers** can watch live streams with near-real-time latency (a few seconds is acceptable) - **Live chat** alongside each stream so viewers can interact - Viewers can browse and discover active streams - Support for **stream recording** so content is available as VOD (video on demand) after the stream ends - Multiple quality levels so viewers on slow connections can still watch ### Non-Functional Requirements - Support **millions of concurrent viewers** on a single popular stream - End-to-end latency under **10 seconds** from broadcaster to viewer - **High availability** — streams should not drop mid-broadcast - The system should scale horizontally as viewership grows - Chat should feel real-time (sub-second delivery for most users) The key insight here is that live streaming is fundamentally a write-once, read-millions problem. One broadcaster produces content that millions consume simultaneously. ### Out of Scope - Payment processing and subscription management (mention it, don't design it) - Content moderation ML pipelines - Mobile app specifics - Authentication and authorization details --- ## Step 2: Back-of-the-Envelope Estimation (5 minutes) Let's get a feel for the numbers. This helps you make informed decisions about architecture later. ### Traffic Estimates Daily active users: 50 million. Concurrent viewers at peak: 10 million. Concurrent broadcasters at peak: 100,000. Average stream duration: 2 hours. Average viewers per stream: 100 (but top streams have millions). ### Bandwidth Estimates Video is a bandwidth monster. 1080p runs at 6 Mbps, 720p at 3 Mbps, 480p at 1.5 Mbps, 360p at 0.8 Mbps. Ingest bandwidth (broadcaster to our servers): 100,000 broadcasters x 6 Mbps = 600 Gbps ingest. Egress bandwidth (our servers to viewers): 10 million viewers x 3 Mbps (average) = 30 Tbps egress. That 30 Tbps number is exactly why CDNs exist. No single data center can push that much traffic. ### Storage Estimates If we record all streams: 100,000 streams x 2 hours x 6 Mbps = 540 TB per day of raw video. After transcoding to multiple qualities, roughly 2x that = ~1 PB/day. With 30-day retention for VOD: ~30 PB. --- ## Step 3: High-Level Design (10 minutes) The system is a pipeline: video flows from the broadcaster through several stages before reaching viewers. ### Core Components 1. Video Ingestion Service - Accepts live video from broadcasters via RTMP 2. Transcoding Service - Converts incoming video into multiple resolutions and bitrates 3. CDN - Distributes video chunks to edge servers close to viewers 4. Chat Service - Handles real-time messaging for each stream 5. Stream Discovery and Metadata Service - Manages stream listings and search 6. VOD Service - Records live streams for later playback --- ## Step 4: Deep Dive (20 minutes) ### The Video Pipeline RTMP ingest, chunking into 2-6 second segments, transcoding to multiple renditions, and delivery via HLS adaptive bitrate streaming. ### CDN Architecture Tiered cache architecture with shield/mid-tier layer between origin and edge. Multi-CDN strategy for redundancy and global coverage. ### Live Chat at Scale WebSocket connections with fan-out strategies that scale from direct delivery to message sampling based on viewer count. ### Stream Recording and VOD Asynchronous pipeline that stitches chunks, re-transcodes, generates thumbnails, and stores with lifecycle policies. --- ## Wrap Up Key trade-offs: latency vs quality, consistency vs availability, cost vs performance. The interviewer wants to see you navigate a complex system with many moving parts and make reasonable decisions at each layer.