Estimate Daily Storage for Video Uploads

Problem

A video-sharing platform at YouTube scale accepts user uploads around the clock. Estimate how much new storage the platform must provision per day to hold what users upload, including the transcoded copies it serves and whatever replication a production storage system needs.

Give a single number with its order of magnitude, and be explicit about every assumption. You are not expected to know any internal figure; you are expected to build the number from things you can defend.

• Treat "storage" as bytes written to durable storage, not network bandwidth.
• Ignore thumbnails, comments, metadata and analytics; they are noise next to the video bytes.
• State the answer in petabytes per day and extend it to a yearly figure.

Examples

Reasonable inputs to state up front; the interviewer will usually accept them or hand you a replacement:

• Upload rate: roughly 500 hours of video are uploaded every minute. If you would rather build it from users: about 2.5 billion monthly users, of whom a small fraction upload on a given day, ending with a few million uploads per day at an average of ~8 minutes each. Both routes should land within a factor of two of each other.
• Source quality: the average upload is 1080p at ~8 Mbps. Some are 4K at 35-45 Mbps, many are 720p phone clips at 4-5 Mbps; 8 Mbps is a fair blended figure.
• Transcoding: every upload is re-encoded into a ladder of renditions (144p, 240p, 360p, 480p, 720p, 1080p, and 4K where the source allows). The interviewer will accept that the whole ladder adds up to roughly the size of the original again, because each rung is a fraction of the one above it and the series converges.
• Durability: 3× replication, or erasure coding at ~1.5× overhead if you prefer. Either is fine as long as you say which.

Hints

1. Convert "hours per minute" into "hours per day" first; the unit conversion is where most people drop a factor of 60.
2. Turn a bitrate in megabits per second into bytes per hour before you multiply by hours. 8 Mbps is 1 MB/s, which makes the arithmetic clean.
3. The original upload is not the only copy. Ask what else gets written: renditions and replicas both multiply the number.