Hotstar’s 25M-viewer play favors failures over auto-scaling
Hotstar (now Disney+ Hotstar) presented its architecture for scaling to 25 million concurrent viewers at AWS re:Invent 2019, detailing how it handled major live events like the Indian Premier League (IPL) and Cricket World Cup. The presentation outlined their approach to battle-tested scaling strategies, preparing for failures, and implementing graceful degradation instead of relying on auto-scaling.
Key Takeaways
- Hotstar said it reached a 25.3M concurrency pattern, with 16.2M and 13.9M also shown in the presentation.
- The company cited major live events including IPL 2019 Final, Super Bowl, and World Cup India vs New Zealand as scale references.
- Its load-generation architecture spanned multiple AWS regions: us-east-1, us-west-1, eu-west-1, and eu-west-2.
- Hotstar said it does not use auto scaling for this workload.
- The operating model includes turning off non-critical services while keeping P0 services always up and degrading gracefully.
Why It Matters
Hotstar’s presentation shows how a large streaming service thinks about live-event reliability: not by assuming autoscaling will catch every spike, but by planning for failure and preserving core playback paths. That matters for any platform carrying sports or other high-concurrency events, where the user journey has to be mapped service by service. The AWS framing also points to multi-region load generation as part of readiness, not just production capacity. What to watch next is whether Hotstar continues to publish concrete concurrency numbers for events like IPL and Cricket World Cup, since those are the clearest markers of how this approach holds up under load.
Read full article at d1.awsstatic.com