Service Time
The time a resource actively spends processing a unit of work, excluding time spent waiting in a queue before service begins.
Separating waiting from execution is essential when diagnosing latency. Service time is the active processing interval once a request reaches the CPU, GPU, database, storage device, or another execution resource.
Two systems with the same service-time distribution may still have very different end-to-end latency because their queueing behavior differs. Capacity analysis therefore needs the distribution and its variability, not only a mean. Capacity engineering for real-time speech recognition connects this distinction to sustained throughput.