Media & Entertainment

ABR dynamically adjusts video quality based on network conditions. HLS and DASH protocols segment video into small chunks (2-10 seconds) at multiple quality levels. The player monitors buffer health and bandwidth, switching to higher/lower quality segments seamlessly. Key components: manifest file listing available qualities, player ABR algorithm (throughput-based, buffer-based, or hybrid), and segment-aligned keyframes for smooth switching.

RTB is an auction for ad impressions in real-time. When user loads page: 1) Publisher sends bid request to SSP, 2) SSP broadcasts to multiple DSPs via OpenRTB, 3) DSPs evaluate against campaigns and return bids, 4) SSP runs second-price auction, 5) Winner's ad is served. Entire process must complete in ~100ms. DSPs use ML to predict conversion probability and calculate optimal bid. Infrastructure requires global presence, pre-computed user segments, and cached campaign data.

Hybrid approach combining: 1) Collaborative filtering - find users with similar watch history, recommend what they watched, 2) Content-based - analyze content features (genre, actors, director), recommend similar content, 3) Contextual - time of day, device type, recent activity. Architecture: event collection pipeline (Kafka), feature store (Redis/DynamoDB), ML models (matrix factorization, neural networks), A/B testing framework. Optimize for diversity (avoid filter bubbles), freshness, and business metrics (watch time, retention).

Key challenges: 1) Latency - sub-100ms round-trip required, use UDP, client-side prediction, server reconciliation, 2) Scale - millions concurrent users need fleet management, matchmaking by region, 3) Cheating - server-authoritative game logic, anti-cheat systems, behavior analysis, 4) State sync - deterministic simulation, delta compression, interpolation/extrapolation, 5) Reliability - graceful degradation, reconnection handling, save state persistence. Use dedicated game servers (not peer-to-peer) for competitive games.

CDNs cache content at edge locations close to users: 1) Origin shield reduces load on origin, 2) Cache warming pre-populates popular content, 3) Multi-CDN strategies for redundancy and optimization, 4) Token authentication prevents hotlinking, 5) Geo-restrictions enforce licensing, 6) Chunked transfer enables seeking without full download. Advanced: predictive pre-fetching, edge compute for personalization, QUIC/HTTP3 for faster connection establishment.

Multi-DRM serves different DRM systems to different devices: Widevine (Chrome, Android), FairPlay (Safari, iOS), PlayReady (Edge, Windows). Implementation: 1) Encrypt content once with Common Encryption (CENC), 2) Generate encryption keys stored securely, 3) License server validates user entitlement, returns content keys, 4) Player requests license, decrypts content in protected memory. Security levels: L1 (hardware TEE), L3 (software-only). Also need: offline playback, concurrent stream limits, output protection.

Multi-layered approach: 1) Upload-time scanning with ML (nudity, violence, hate speech classifiers), 2) Hash matching against known bad content (PhotoDNA, CSAM databases), 3) User reporting pipeline, 4) Human review for edge cases and appeals. Prioritization: severity (child safety first), virality (fast-spreading content), context (news vs. promotion). Infrastructure: async processing queues, moderator tooling, quality assurance, moderator wellbeing. Also: proactive detection, coordinated inauthentic behavior detection, transparency reporting.

User experience: rebuffer rate, video start time, bitrate quality, playback failures, stream starts per user. Engagement: watch time, completion rate, sessions per user, content diversity consumed. Business: subscriber churn, conversion rate, ARPU. Infrastructure: CDN hit ratio, origin load, encoding queue depth, DRM license latency. Alerting on: spike in errors, geographic issues, device-specific problems. Use real-user monitoring (RUM) and synthetic monitoring for comprehensive coverage.