AI & Technical question

Design a system to store large videos.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Infrastructure design for large binary object storage at scale, separating metadata from blob storage and delivery.

Clarify requirements: scale of uploads per day, access patterns, write-once-read-many, and durability needs.
Chunk large video files into smaller segments for parallel upload and download and resumable uploads on failure.
Use object storage for the video bytes, with a separate metadata database, title, owner, encoding status, pointing to storage locations.
Address durability: replicate chunks across multiple data centers or availability zones to survive hardware failure.
Address delivery: use a CDN to cache and serve popular segments close to viewers, reducing origin load and latency.
Define success as upload success rate, including resumability after network failure, and playback start latency at scale.

Separates metadata storage from blob storage explicitly, a core, correct systems-design distinction.
Names chunking plus resumable uploads as a concrete answer to the real-world problem of large-file upload failures.
Addresses delivery via CDN, not just storage, recognizing 'store' implicitly includes serving it back efficiently.

Only discussing storage and ignoring resumable upload handling for large files, a very real failure mode.
Ignoring delivery and CDN, treating this as a pure storage problem with no serving-side consideration.

How would you handle a video still being uploaded when a user tries to watch it?
How would you estimate the storage cost at scale?