01What is the catalog shape — user uploads or licensed titles?
Licensed and curated: thousands of titles, not billions of uploads — which flips the design from ingest-heavy (a UGC video platform) to delivery-heavy (this question).
Premium preview
Netflix
Design Netflix’s video delivery system — stream a licensed, curated catalog to tens of millions of concurrent viewers worldwide with minimal buffering and consistent quality.
00
Practice checkpoints
The requirements are open as a taste. From the numbers onward, the full guide opens in the app.
- 01Clarify scope
- 02Requirements + scale
- 03API + data modelUnlocks in the app
- 04Draw architectureUnlocks in the app
- 05Deep diveUnlocks in the app
- 06Trade-off decisionUnlocks in the app
01
Requirements that shape the design
Do not only state requirements. Ask for them. Each card pairs the design constraint with a clarification question you can say out loud before drawing the architecture.
Functional requirements
02How fast must play start, everywhere?
About a second to first frame, globally — which is only possible when the bytes are already near the viewer before they press play.
03Does watching follow the user across devices?
Yes: resume points sync continuously; stopping on one device and continuing on another at the same second is a core promise, not a nicety.
04What does a global premiere demand?
A scheduled title must stream flawlessly at launch minute in every region — pre-positioning to edges happens ahead of time, on a schedule, not on demand.
05One encode per title, or per device class?
One per-title ladder covering all devices: resolutions and bitrates optimized per title (animation compresses differently than action), served adaptively.
06Subtitles, audio languages, regional variants?
Tracks are catalog metadata delivered beside the stream; region licensing decides catalog visibility, playback machinery stays identical.
Out of scopeRecommendation ranking (the Recommendation System question) · Content production and licensing workflows · Account/billing and password sharing enforcement
Non-functional requirements
01What single metric defines success?
Rebuffer ratio: fractions of a percent of total watch time — start time matters, but mid-stream stalls are what cancel subscriptions.
02Where do the bytes actually come from?
Mostly from edge caches inside ISP networks (Open Connect style): the popular catalog is pre-positioned overnight, so peak traffic barely touches the backbone.
03An edge cluster dies mid-evening — what do viewers see?
Nothing: clients re-resolve to the next-best edge or another CDN mid-stream; a quality dip is acceptable, a stall is not.
04Can the whole catalog live at every edge?
The popular head (which serves the vast majority of hours) yes; the long tail streams from regional origins through the cache hierarchy — misses are planned, not accidents.
05What load do we size for?
Peak concurrent evening viewing per region — tens of millions of concurrent streams; averages are irrelevant to capacity.
Keep asking — the interview is a conversation
Real interviews probe far more than a tidy list. These are the scope questions that separate candidates who interrogate the problem from those who recite it.
- Licensed catalog or user uploads? The whole design pivots on this.
- What rebuffer target and start-time target define quality?
- Do we control edge hardware inside ISPs, or ride commercial CDNs?
- How predictable is demand — do premieres dominate capacity planning?
- Is offline download in scope?
01
Unlock the full playbook for Netflix
Numbers, architecture diagram, API and data model, deep dives, expected topics, self-check, whiteboard starter, and common mistakes unlock inside the app.
Numbers that force architecture decisions
Locked in the app
Architecture path
Locked in the app
API and data model
Locked in the app
Deep dive directions
Locked in the app
+Series