YouTube / Netflix Streaming

ABR, CDN, transcoding, hot/cold — the egress, fan-out, and popularity Pareto.

SavedSaved on this device

The problem

ABR, CDN, transcoding, hot/cold — the egress, fan-out, and popularity Pareto.

Build a video streaming platform like YouTube or Netflix. Users press play and expect a 4 K stream to start within two seconds; creators upload 50 GB mezzanines and expect the full ABR ladder published within minutes. Almost every byte the system serves is read; almost every byte it stores was produced by transcoding. The architectural pressure is not QPS — it is egress (you are 15% of the US internet at peak), fan-out (one source mezzanine becomes 250 K encode jobs across rungs × codecs × shots), and the popularity Pareto (the top 1% of catalogue serves 50% of bytes; the bottom 50% serves <0.1%).

The reference design below is what an SRE staff team at Netflix-class scale would actually run a five-year incident retro against — not the 45-minute whiteboard sketch. Every node has load-bearing internals; every edge declares wire semantics; every chaos card is grounded in a cited real-world incident. Focus is on the four pillars: ABR, CDN, transcoding, and hot/cold tiering. Comments / search / recommendations / ads are explicitly out of scope.

Reading: Netflix Tech Blog — Per-Title Encode Optimization (2015) · Netflix Tech Blog — Optimized Shot-Based Encodes (2018) · Netflix Tech Blog — Rebuilding Video Pipeline with Microservices (Cosmos, 2023) · Netflix Tech Blog — Serving 100 Gbps from a single Open Connect Appliance · Netflix Open Connect — Overview whitepaper + Fill Patterns docs · Google / YouTube — Reimagining YouTube Video Infrastructure (Argos VCU, 2021) · Apple — Enabling Low-Latency HLS · Hotstar Engineering — How we scaled to 25M concurrent for IPL · AWS — CloudFront Origin Shield in Multi-CDN Deployments · BBA paper — Stanford SIGCOMM 2014 (Buffer-Based ABR) · Pensieve — MIT SIGCOMM 2017 (RL-based ABR)

ABR (HLS / DASH / CMAF)
multi-tier CDN
per-title / per-shot encoding
DRM (Widevine / FairPlay / PlayReady)
hot/cold tiering
transcode fan-out

01Clarifications

What would you ask before drawing a single box?

Ambiguity you would resolve with the interviewer: scope, scale, who uses it, what counts as done.

About YouTube / Netflix Streaming

ABR, CDN, transcoding, hot/cold — the egress, fan-out, and popularity Pareto.

Difficulty: advanced
Time: about 60 minutes
Stages: 10
Topic: Object Storage, File Sync & Media Delivery

How this problem is worked

Ten stages, from the questions you would ask an interviewer to the trade-offs you would defend. Each asks one question, and the simulator runs the architecture you draw against the requirements you wrote.

01ClarificationsWhat would you ask before drawing a single box?
02Functional reqsWhat must this system actually do?
03Non-functionalWhat must it promise about speed, uptime and correctness?
04Capacity estimationHow much load and data does this have to hold?
05API designWhat does the outside world call, and what comes back?
06Data modelWhat gets stored, and what is it looked up by?
07Use-case breakdownHow does each requirement actually get served?
08High-level designWhich components handle a request, and in what order?
09Deep divesWhich part breaks first, and what do you do about it?
10Trade-offsWhat did this design cost, and what breaks at 10×?

Primary sources for this problem

Netflix Tech Blog — Per-Title Encode Optimization (2015)
Netflix Tech Blog — Optimized Shot-Based Encodes (2018)
Netflix Tech Blog — Rebuilding Video Pipeline with Microservices (Cosmos, 2023)
Netflix Tech Blog — Serving 100 Gbps from a single Open Connect Appliance
Netflix Open Connect — Overview whitepaper + Fill Patterns docs
Google / YouTube — Reimagining YouTube Video Infrastructure (Argos VCU, 2021)
Apple — Enabling Low-Latency HLS
Hotstar Engineering — How we scaled to 25M concurrent for IPL
AWS — CloudFront Origin Shield in Multi-CDN Deployments
BBA paper — Stanford SIGCOMM 2014 (Buffer-Based ABR)
Pensieve — MIT SIGCOMM 2017 (RL-based ABR)

Build the primitives this design leans on

Each one is an animated curriculum that constructs the system from scratch.

Build Build a CDNA globally-distributed reverse proxy whose only job is to (a) terminate the user's TCP/TLS milliseconds away and (b) serve a cached origin response so origin never sees the request. Internalize edge caching, anycast, TTL, revalidation, SWR, purge, the Vary footgun, origin shield, bypass, and hit ratio — and the dozen ways to misconfigure each.

More in Object Storage, File Sync & Media Delivery

Durable bytes at rest and bytes in flight: an S3-style object store, a Dropbox-style sync client, and adaptive video streaming.

Browse the full problem catalog, or see what the simulator does and does not model.