← Create New Paste

How to Answer "Design a Video Streaming Platform" in a System Design Interview

4 views
"Design YouTube" or "design Netflix" is one of the most common system design prompts, because it touches storage, encoding, delivery and scale all at once. Here is a structure that keeps you organized under time pressure. 1. Clarify scope first Ask whether you are designing upload, playback, or both. Confirm scale assumptions: for example, 500 million daily viewers, 500 hours of video uploaded per minute, and a global audience. Agree on non-functional goals: smooth playback, low startup latency, and high availability matter more than strict consistency for view counts. 2. Upload pipeline Clients upload raw files to object storage using pre-signed URLs, so your API servers never proxy large blobs. An upload-complete event goes onto a queue. Transcoding workers pick it up, split the video into segments, and encode each segment at several resolutions and bitrates (240p through 4K). Metadata such as title, owner, status and available renditions lives in a database; a relational store sharded by video ID is a reasonable default. 3. Playback and adaptive bitrate Explain HLS or DASH: the player downloads a manifest listing segment URLs per quality level and switches quality as bandwidth changes. This is the single most important concept interviewers want to hear, because it explains how playback stays smooth on bad networks. 4. Content delivery Popular videos are served from a CDN close to the viewer. Mention a tiered approach: hot content cached at edge locations, long-tail content fetched from origin storage on demand. Pre-warming the CDN for an expected viral release is a nice detail. 5. Supporting services View counts can be aggregated asynchronously with a stream processor rather than updated per request. Recommendations, search and comments are separate services; name them, but don't go deep unless asked. 6. Bottlenecks and trade-offs Discuss transcoding cost (encode popular videos in more formats, rare ones lazily), storage growth, hot-key videos, and regional failover. Interviewers reward candidates who pick a trade-off and justify it. Practicing out loud The hardest part is not knowing these pieces; it is laying them out calmly in 45 minutes while someone interrupts with follow-ups. Practicing with a timer and talking through each box helps a lot. Tools like Linkjob AI (https://linkjob.ai) can also give real-time help during live technical interviews, surfacing structure and key points when your mind goes blank. Whatever you use, rehearse the flow until "upload, transcode, manifest, CDN" feels automatic.
Created Nov 15, 2026 19:31 Expires Nov 15, 2027 19:31