CPUytimet

CPUytimet is a theoretical performance metric used in computer science to quantify the amount of time a central processing unit spends performing productive work during a defined observation window. It aims to reflect not only raw execution time but also the effects of memory latency, pipeline stalls, and context-switch delays that occur while the CPU remains active on a workload.

The term is not standardized and is used primarily within research contexts to complement traditional measures such as CPU time and wall-clock time. CPU time sums all time the CPU spends executing user and system code, while wall-clock time measures real elapsed time. CPUytimet, by contrast, seeks to represent the portion of CPU activity that directly advances the workload, potentially incorporating short-term stalls that do not lead to observable progress.

Definition and scope vary, but a common interpretation defines CPUytimet for a workload window as the accumulated time during which the CPU performs productive work, including instruction retirement and short, productive memory operations, while excluding long idle periods caused by power gates, resource contention, or external scheduling delays. Different measurement approaches exist, including hardware performance counters, architectural simulators, and operating-system telemetry, each requiring a clear, workload-specific standard for what counts as productive activity.

Applications of CPUytimet include evaluating scheduling policies, analyzing the efficiency of microarchitectures, and studying energy-performance trade-offs in virtualization environments. Limitations include the lack of universal agreement on its definition, sensitivity to workload and measurement method, and potential overlap with other metrics.

See also: CPU time, CPU utilization, wall-clock time, instructions per cycle.