Streaming Services

ABR streaming encodes content at multiple quality levels (renditions) and lets the client player dynamically switch between them based on available bandwidth. Protocol: Content is split into small segments (2-6 seconds). A manifest file (M3U8 for HLS, MPD for DASH) lists all available renditions with bitrates. The player downloads segments and measures download speed. If bandwidth drops, it switches to a lower quality — preventing buffering. If bandwidth improves, it switches up for better quality. Why necessary: Users have varying network conditions (2G rural India to 5G urban), different devices (mobile to 4K TV), and bandwidth fluctuates. Without ABR, you'd either buffer on slow connections or waste bandwidth on fast ones. Example: JioCinema serves 240p on 2G and 4K HDR on fibre — same content, adaptive quality.

Key architecture: 1) Multi-CDN: Don't rely on one CDN — use Akamai + CloudFront + ISP-level CDN (Jio CDN in India). Intelligent CDN switching based on edge load. 2) Edge caching: Live segments cached at edge for 2-4 seconds. Same segment served to millions without hitting origin. 3) Auto-scaling: Pre-scale infrastructure hours before the event. Kubernetes pods scale based on viewer count predictions. 4) Tiered architecture: Origin (encode) → Mid-tier cache → Edge PoP → ISP cache → Viewer. Each tier reduces load on the upstream. 5) Graceful degradation: If capacity is strained, reduce max quality (serve 720p instead of 4K) rather than drop viewers. 6) Redundancy: Dual encoders, dual CDN paths, multi-region origin. JioCinema's IPL approach: Pre-positioned content at ISP level, Jio network-level caching, dedicated CDN for Jio users.

Netflix's recommendation is a multi-stage system: 1) Candidate Generation: From a catalogue of thousands, narrow to ~1000 candidates per user using collaborative filtering (users with similar taste) and content-based signals (genre, actors, mood). 2) Ranking: A deep learning model scores each candidate — features include user watch history, time of day, device, trending signals. 3) Re-ranking: Apply business rules — promote new releases, ensure diversity (don't show all thrillers), contractual obligations. 4) Row Generation: Home screen is organised in rows ('Because you watched Breaking Bad', 'Trending in India'). Each row has its own model. Row order is also personalised. 5) Artwork Personalisation: Even the thumbnail image is personalised — if you watch comedies, you see the funny scene; if you watch action, you see the action scene from the same title. 80% of what users watch comes from recommendations (not search). The system processes billions of events daily to keep models fresh.

DRM (Digital Rights Management) prevents unauthorised copying and distribution of content. How it works: Content is encrypted using AES-128. The decryption key is stored on a license server. When a user plays content, the player requests a license (decryption key) — the server checks if the user has a valid subscription and returns the key. The key is handled in a secure environment (TEE/hardware) and never exposed to the user. Multi-DRM: Different platforms require different DRM systems — Widevine (Android, Chrome), FairPlay (iOS, Safari), PlayReady (Windows, Edge). CENC (Common Encryption) standard allows encrypting content once and using any DRM to deliver the key. A multi-DRM service (like PallyCon or BuyDRM) manages all three DRM systems. Security levels: L1 (hardware-secured, allows 4K/HDR) vs L3 (software-only, limited to 720p on many platforms).

India's language diversity is a unique streaming challenge: 1) Multi-language audio: Same content available in Hindi, Tamil, Telugu, Kannada, etc. — stored as separate audio tracks, selected by user preference. 2) Subtitles: WebVTT/SRT subtitle tracks in 10+ languages. Auto-generated using ASR (speech recognition) and machine translation. 3) Regional content: Separate content libraries for each language. ZEE5 has 12 language-specific catalogues. JioCinema has regional originals. 4) Recommendation localisation: Models consider language preference as a strong signal. A Tamil-speaking user in Chennai sees Tamil content prominently. 5) UI localisation: App interface in local language. 6) Search: Multi-script search — user can type in Devanagari, Tamil script, or transliterated English and find the same content. This is a competitive advantage for Indian platforms over Netflix which historically struggled with regional content.