Simulation scenarios
Each scenario replays a real-world event against the cluster. Open the dashboard to watch the optimizer respond.
A push notification lands and Chennai climbs from 230 to 640 rps over 15 seconds — past the region's 600 rps cap. The forecaster should flag the breach early and pre-warm inference-api on edge-chennai before p95 crosses the 20ms SLA.
A flash sale goes live with zero lead time: Chennai spikes to 900 rps in 5 seconds. Too fast for a clean pre-warm — watch how the optimizer reacts when the forecast lags the traffic instead of leading it.
The Mumbai PoP loses grid power mid-operation. Reliability drops to zero and whatever runs there must be evacuated to core before queues back up — zero dropped requests is the bar.
Traffic falls to the overnight floor for five minutes. Edge nodes burn watts for headroom nobody needs — the optimizer should consolidate workloads back onto core and cut the energy draw by roughly 40%.
Chennai surges to 700 rps, then edge-chennai — the node best placed to absorb it — dies eight seconds in. The optimizer has to serve the surge from core alone. The hardest scenario in the set.
Grid power returns and demand normalizes: every node revives and the demand curve resets to baseline. Use this to bring the cluster back to a clean steady state between runs.