Push a server — or a team — toward 100% busy and the wait doesn't climb steadily, it runs off a cliff. Queueing theory (M/M/c) says how steep: at 85% load a lone worker already makes every job wait about as long as the work itself, and by 95% it's nineteen times longer. Add workers sharing one queue and the cliff slides right — several pooling their idle time wait far less than each running alone. Set a load and read off the wait, or set the wait you'll tolerate and read off the busy ceiling.
Drag the curve — or focus it and use arrow keys — to set how busy you run.
At 85% busy, every job waits 5.7× as long as the work itself before it even starts.
| Busy | Queue wait |
|---|---|
| 50% | 4m |
| 75% | 12m+8m |
| 85%you | 22m 40s+10m 40s |
| 90% | 36m+13m 20s |
| 95% | 1h 16m+40m |
| 99% | 6h 36m+5h 20m |
Read down the right-hand column: every step up in load costs more wait than the one before it. The jump from 95% to 99% busy alone dwarfs the entire climb beneath it. Add workers sharing one queue and the whole curve slides right — several pooling their idle time absorb bursts that would swamp any one of them alone. That's why you keep some slack in reserve, and why a shared queue beats a queue each.