Queueing delay
Queueing delay is the time a request spends waiting for service at a specified resource rather than receiving service there.
For one nonpreemptive stage, Q = service start - arrival. With preemption, waiting can recur after processing begins; elapsed time until the first start is then only part of the wait. [1]
Queue length counts requests, while queueing delay measures time. Little's law uses matching boundaries: an occupancy that includes requests being processed needs residence time, not waiting time alone. [2]
In the existing synthetic FIFO trace, the second independent arrival waits 101 - 10 = 91 ms and then takes 1 ms to complete. Its 92 ms response therefore does not mean 92 ms of processing. [3]
In a constructed preempt-resume example, A waits 4 ms after first starting and B waits zero. Across both completed requests, mean total waiting is 2 ms, not 4 ms. Counting distinct requests and summing all their waiting intervals is different from counting queue visits. [3]
Relations
- related
- Service time, Storage latency, Completion latency (fio), Little's law, Pollaczek-Khinchine mean formula, Kingman approximation, First-come-first-served scheduling, Head-of-line blocking, Burke's theorem, Tandem queue, Request slowdown