Scaling hotstar.com for 25 million concurrent viewers
summary
Gaurav Kamboj describes Hotstar’s approach to serving 25.3 million concurrent viewers, explaining why standard autoscaling is insufficient for their live-video workloads and how they instead run full-scale gamedays. He covers their custom load-testing tool Hulk used to rehearse peak traffic and the use of chaos engineering practices to uncover and fix scalability weaknesses.
AI-written summary · about
chapters (13)
- 0:00Start
- 2:59Concurrency Pattern
- 6:03About hotstar
- 6:58Let's talk about Scale!
- 8:02Why is that a big number?
- 9:13Game Day
- 13:39Race against time!
- 15:07Why we don't use autoscaling..
- 19:08Battle tested scaling strategy..
- 20:37Ingredients for actual Chaos at hotstar
- 25:33What are we looking for?
- 27:11Panic Mode..
- 29:51Key Takeaways..
related talks
29mAnalysing high throughput data in real-time
47mScaling: from 0 to 25 million users
08h 26mLivestream
29mHow Seven delivers large-scale streaming success
13mScaling for a Pandemic: How We Keep Ahead of Demand for Google Meet during COVID-19
22mScaling Culture of Resiliency in the Enterprise at Charter Communications
19mEverything I learned about dynamic scaling of cloud apps.
08mGetting Creative with HPA Scaling in Kubernetes