The Brownout Dial

Controls

Arrival rate λ

How many requests per second arrive. Push it past capacity (servers × 3 req/s each) to trigger overload.

Servers c

How many workers serve requests in parallel. Capacity = servers × 3 req/s.

Client deadline

Maximum elapsed time from arrival to completion, including queue wait and service. A request that completes after this deadline still consumes service and counts as throughput, but not useful goodput.

Shedding

Admission control: reject new arrivals when requests being served or waiting reach the admission limit. The default limit uses worker capacity and the client deadline to keep the queue short.

Retries

Rejected or timed-out requests come back after a short delay, adding to the arrival load.

Retry budget

A token bucket limits retry re-entry over time: retries spend tokens and the bucket refills. It does not cap concurrent requests.

Run rescue sequence

Sets load to 3.5× service capacity, then stages overload, shedding, a retry storm, and a retry budget. With normal motion, stages run at five-second intervals and can be stopped. With reduced motion, this same button changes to a named Advance rescue action for each stage, then offers a restart. Change a control to take over.

Stop rescue sequence

Only shown during timed playback: cancel pending stages and keep the settings already reached. Change a control to take over.

Reseed

Builds a fresh chart and queue with the current settings and a new visible seed, so you can compare another reproducible run.

Step model

When reduced motion is enabled, advances the simulation in four paced frames while automatic animation stays paused.

What to watch for

• With shedding OFF and load high, watch Goodput climb, peak, then fall — even as Utilisation stays pinned near 100%.
• With shedding OFF, a full waiting room can still reject arrivals; those are capacity drops, not policy fast-rejects.
• Turn Shedding ON at the same load. Existing requests stay in the queue; give the backlog time to respond before comparing Goodput. Rejected arrivals still include both policy rejects and full-queue drops.
• With Retries ON and no shedding, the collapse can worsen — queue-full rejects and timed-out work retry, adding MORE load to an already-overloaded system.

Why it happens

The queue processes admitted requests in FIFO order. A request can still consume service after its client deadline expires; that completion adds throughput but not goodput. The finite waiting room can reject an arrival at capacity even when shedding is off; the shedding control adds an earlier, deliberate admission threshold. The rejection metric combines both causes. Changing a control does not clear queued requests or chart history; Reseed starts a fresh run with the current settings. Its capacity line is a service-rate reference, not an SLO; a real SLO also needs a latency or error objective.