JioHotstar System Design: Why Autoscaling Fails at IPL Toss @KGPTalkie
JioHotstar System Design: Why Autoscaling Fails at IPL Toss  @KGPTalkie
Uploaded October 2026 | Updated November 2026, 3 days ago
IPL final, the toss: crores of people open JioHotstar at the same time, and autoscaling won't save you.

System Design Interview, Day 002. The question: design it so it doesn't go down. Most candidates say "turn on autoscaling", and that's where the round slips away. Do the maths: ten lakh people open the app every minute before the toss, and a new server takes about three minutes to start. Thirty lakh people hit the old servers first.

- Fix 1: scale before the match. Forecast the peak from the last final and load tests, add a buffer, warm servers, databases and CDN capacity before the toss. Scale on concurrent viewers, not CPU
- Fix 2: the video pipeline. One stream from the stadium encoder, a transcoder makes qualities from 1080p down to 360p, a packager cuts six-second pieces at the same moments, one playlist per quality
- Fix 3: an origin shield. In live every piece is new, so every edge misses at once. A shield layer fetches each piece from your server only once
- Fix 4: adaptive bitrate. The phone measures its speed on every piece and steps quality down or up. The price: you watch a few seconds behind live
- The trap: video comes from the CDN, but login, the subscription check and the DRM licence still hit your servers at the toss. Cache the subscription check, spread token refreshes out
- Fix 5: a panic mode. A big wicket, millions press back. Feature flags switch off recommendations and watch history, circuit breakers skip slow services, the gateway sheds load. Video, login and payments keep working
- Fix 6: jitter for the ad break. Each phone waits a random half second to three seconds; retries use exponential backoff

What the interviewer wanted: more than autoscaling. Scale ahead, an origin shield, adaptive bitrate, a panic mode, and jitter.

Next System Design Interview episode: you send 500 rupees on UPI, the money leaves your account, but the app says failed. Where did it go? Comment: which moment is the hardest, the toss, the wicket, or the ad break?

kgptalkie.com

#shorts #SystemDesign #SystemDesignInterview #IPL #JioHotstar #LiveStreaming #CDN #Autoscaling #InterviewPrep #KGPTalkie
JioHotstar System Design: Why Autoscaling Fails at IPL Toss13 Gen AI Interview Preparation: What is MLM Maked Language ModellingVideo Encoding and CDN Explained: 20 TB, 1 TB From ServerNLP Projects 2 - Build IMDB Sentiment Classification Application with Streamlit | NLP TutorialClaude + Blender 🤯 This Changes 3D Design Forever (Full Tutorial)Fable 5.2 vs GPT-6 Astra: Same Score, 4x Cheaper?NVIDIA NIM Free API Key: Free Models on build.nvidia.com (Nemotron, DeepSeek)04 Gen AI Interview Preparation: How to Design Production Ready RAG Systems #ai #coding #interviewOpenClaw Use Cases - Build a Free Stock Researcher AI Bot in Telegram (2026)How to Install Docker on Ubuntu | Install Docker on Linux | Docker InstallationClaude Fable 5.2 vs Opus 5.5: Five 3D Games, One Shot EachOllama Was 2× Faster Than llama.cpp. Heres Why
Laxmi Kant | KGP Talkie |

JioHotstar System Design: Why Autoscaling Fails at IPL Toss

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER