Netflix — Mock Interview

Hard

A planet-scale video-on-demand service: stream a curated, pre-encoded catalog to tens of millions of concurrent viewers from the edge.

Interviewer

You're designing Netflix for a live system-design interview. Members browse a curated catalog, get personalized recommendations, and stream titles on any device with adaptive quality and resume. This is not a user-upload site — content arrives from studios and is pre-encoded offline into an adaptive-bitrate (ABR) ladder long before playback.

The defining challenge is delivery at scale: tens of millions of concurrent streams add up to hundreds of terabits per second of egress that no central datacenter can serve — the bytes must come from caches inside ISPs (Open Connect).

Scale assumptions:

~250M subscribers; at prime time ~20% (~50M) streaming simultaneously.
Average stream ~5 Mbps → peak egress on the order of 250 Tbps; ~450 PB delivered per peak day.
Catalog ~20,000 titles; encoded ABR-ladder storage ~2 PB.
Targets: start < 2 s, rebuffer < 0.5%, playback availability 99.99%.

We'll go through requirements, estimation, architecture, the CDN/ABR delivery deep dive, and trade-offs. Think out loud and state your assumptions.

The interviewer grades each stage against a hidden rubric and asks a probing follow-up, mirroring a real system design round. Sessions are capped at 30 turns.