System model (the service)
System diagram (queueing network)
Workload model (the demand)
A performance requirement is meaningful only against a stated workload.
The model changed since this run. Simulate again.
Summary
Little’s Law
Resource utilization
Queueing
Four experiments, in order. Each sets the laboratory up for you; the point is the reasoning, not the clicking. Predict before you run — the prediction is the experiment.
Model assumptions
- Arrivals are stochastic (Poisson) or exactly spaced (constant), as selected.
- Service times are exponential unless a preset states otherwise; stations draw independently.
- Requests follow a fixed route through the stations; there is no conditional routing.
- Queues are unbounded unless a capacity is given. A full queue refuses work rather than growing.
- Timeouts, retries and cancellation are not modeled. Retry amplification therefore cannot appear here, though it is real.
- A resource is released when its station completes; nothing is held across downstream calls.
- The first 10 seconds are discarded as warm-up. Reported figures are measurements of the remaining window, not steady-state limits.
- A cache hit is modeled probabilistically as a route that skips the database. No cache memory, freshness or invalidation cost is represented.
Simulation results describe the model you specified, not necessarily the deployed system.