CPU and Processor Time Counter
Read this introduction to computer performance and information about computing processor time. In most programming languages, there is a process to measure elapsed time, such as time() in C. By subtracting the time at the beginning of a process from the time at the end you get the total time for a particular operation. Usually short operations are put in a loop that repeats the operation a sufficient number of times to get an accurate measurement.
About
CPU time: The count of cycles, also known as clockticks, forms the basis for measuring how long a program takes to execute.
Definition
How do you measure Execution Time?
\( \begin{array}{rll} && \text{Kernel} && \text{Time} \\ + &&\text{User} && \text{Time} \\ \hline \\ && \text{Total} && \text{Time} \end{array} \)
| CPU Time | Running which code |
|---|---|
| Kernel (Sys) Time | Operating system code - the time of CPU spent on the kernel (system) code only within the given process - other processes and the time when our process is blocked are not included. |
| User Time | Time CPU spends running the program code (ie CPU time spent on non-kernel (user-mode) code only within the given process – so other processes and the time when our process is blocked are not included.). |
| Total time | User + Kernel |
| Real Time | Wall clock (total elapsed time) (can be lower than total time because of parallelism or longer because of wait) - user-perceived time it took to execute the command – from the start to the end of the call, including time slices used by other processes and the time when our process is blocked (e.g. I/O waiting) |
Process Explorer
To see CPU time, you can look at the performance tab of a process in Process Explorer:
Red in the CPU usage graph indicates CPU usage in kernel-mode whereas green is the sum of kernel-mode and user-mode execution.
A data collector set can be configured via logman.exe to log the “% Processor Time” counter in the “Processor Information” object for this purpose.
Formula
Every conventional processor has a clock with a fixed cycle time (or clock rate). At every CPU cycle, an instruction is executed.
\( \text{CPU Time} = \text{CPU cycles executed} * \text{Cycle times} \)
\( \text{CPU cycles} = \text{Instructions executed} * \text{Average Clock Cycles per Instruction (CPI)} \)
Putting it all together:
\( \text{CPU Execution Time} = \text{Instructions count} * \text{CPI} * \text{Clock Cycle Time} \)
where:
\(\begin{array}{rll} \text{Instructions Count} && = && \frac{\displaystyle Instructions}{\displaystyle Programs} \\ \text{CPI} && = && \frac{\displaystyle Cycles}{\displaystyle Instruction} \\ \text{Clock Cycle Time} &&= && \frac{\displaystyle Seconds}{\displaystyle Cycle} \end{array}\)
Note:
-
CPI is somewhat artificial (since it is computed from the other numbers) but it seems to be intuitive and useful.
-
Use dynamic instruction count (#instructions executed), not static (#instructions in compiled code)
Performance
Performance is the response time distributed through the request lifetime.
Busy vs Wait
The busy vs. wait percentage shown in monitoring tool is generally:
\( \text{CPU Busy vs Wait Event} = \frac{\text{User} + \text{Kernel}}{\text{Wallclock Time}} \)
CPU Performance
\( \text{CPU Performance} = \frac{1}{\text{Total CPU Time}} \)
System Performance
\( \text{System Performance} = \frac{1}{\text{Wallclock Time}} \)
Units (MHz to ns)
-
Rate is often measured in MHz (millions of cycles per second)
-
Time is often measured in ns (nanoseconds)
\(\begin{array}{rl} {X} \mbox{ MHz } && = && \frac{\displaystyle 1000}{\displaystyle X} \mbox{ ns} \\ 500 \text{ MHz} && \approx && 2 \text{ ns clock} \end{array}\)
Source: DataCademia, https://datacadamia.com/counter/resource/system/cpu/time
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 License.