Back-of-envelope sizing of YouTube's global server fleet

Estimate how many servers YouTube needs to run. Treat it as a back-of-envelope exercise: break the problem into parts, give each assumption a round number, work through the arithmetic, and check whether your final figure is plausible. ```hint Find what limits each machine Watching, uploading and storing video stress different resources. For each kind of work, ask what runs out first on a single server: processor, network or disk. ``` ### Clarifying Questions - Is this for today's global traffic, or for one region or a future year? - Which work counts: only delivering video, or also receiving and processing uploads, storage, and the services behind recommendations, search and ads? - Do cache machines placed inside internet providers' networks count, and should YouTube's share of shared Google infrastructure be included? - Should the fleet be sized for average load or for peak load with spare capacity for failures? ### What a Strong Answer Covers - A clear breakdown of YouTube's workloads, with an equation for each - Round-number assumptions for users, watch time, uploads and per-server capacity, each labeled as an assumption, with the limiting resource named - Correct arithmetic and units, including peak load and redundancy - A sanity check and a view of which assumptions move the answer most ### Follow-up Questions - Which single assumption would you research first to narrow your range, and why? - How would the estimate change if short vertical videos became most of what people watch? - If dedicated video-encoding chips replaced general-purpose servers for processing uploads, which part of your fleet would change and in which direction?