Skip to main content

SLO Simulation

"Find the exact load where your promise breaks — before your users find it for you."

Key Insight​

This project sets a concrete SLO — for example, p95 TTFT under 500 ms — then steadily raises the request arrival rate until the system can no longer meet it, pinpointing the bottleneck that gives out first.

Why This Matters​

Knowing the precise load at which your promise breaks turns a vague "it feels slow sometimes" into a number you can plan around — and tells you whether to fix the scheduler, add GPUs, or shrink the model.