Reactive Scaling
The autoscaler changes capacity only after queue pressure is already visible. Traffic moves first; capacity follows later.
Interactive Systems Explainers
Autoscaling Playground
See how autoscaling reacts to traffic spikes.
Key Observations
What to Notice
Aggressive Scaling
Fast reactions can overshoot. Extra capacity arrives in bursts, then disappears as the system swings back.
Conservative Scaling
Slow decisions avoid thrashing, but the queue stays hot longer while the system catches up.
Horizontal vs Vertical Scaling
Horizontal scaling spreads work across more workers. Vertical scaling makes one worker stronger while the bottleneck stays central.
Real-World Context
Autoscaling turns capacity into an operational policy. It shows up in worker fleets, container platforms, serverless functions, stream processors, and job systems where demand changes faster than humans can react.
The tuning problem is deciding which signal deserves trust: CPU, queue depth, request rate, latency, or a blend. Delayed reactions create backlog; overreactions waste money and cause churn. Each choice trades cost, responsiveness, and stability. Bad tuning also makes capacity planning noisy, because symptoms move between cost and latency.
Interactive Systems Explainers