Hotstar and others often don’t solely rely on traditional auto-scaling methods for critical events because traditional auto-scaling can be too slow. For instance, it can take 200-300 seconds for 200 servers to spin up, which is too long for “tsunami traffic” that requires rapid scaling. Instead, Hotstar uses proactive automated scaling algorithms based on historical data and anticipated demand to provision machines before the rush. They also employ strategies like “preparing for three times of peak” and obsessive load testing with “homegrown” load generators to ensure readiness.