| 1 | <?xml version='1.0' encoding='UTF-8'?>
|
|---|
| 2 | <!DOCTYPE topic PUBLIC "-//OASIS//DTD DITA Topic//EN" "topic.dtd">
|
|---|
| 3 | <topic xml:lang="en-us" id="ts_host-freq-scaling">
|
|---|
| 4 | <title>Frequency Scaling Effect on CPU Usage</title>
|
|---|
| 5 |
|
|---|
| 6 | <body>
|
|---|
| 7 | <p>
|
|---|
| 8 | On some hardware platforms and operating systems, CPU frequency
|
|---|
| 9 | scaling may cause CPU usage reporting to be highly misleading.
|
|---|
| 10 | This happens in situations when the host CPU load is significant
|
|---|
| 11 | but not heavy, such as between 15% to 30% of the maximum.
|
|---|
| 12 | </p>
|
|---|
| 13 | <p>
|
|---|
| 14 | Most operating systems determine CPU usage in terms of time
|
|---|
| 15 | spent, measuring for example how many nanoseconds the systems or
|
|---|
| 16 | a process was active within one second. However, in order to
|
|---|
| 17 | save energy, systems can significantly scale down CPU speed when
|
|---|
| 18 | the system is not fully loaded. When the CPU is running at for
|
|---|
| 19 | example one half of its maximum speed, the same number of
|
|---|
| 20 | instructions will take roughly twice as long to execute compared
|
|---|
| 21 | to running at full speed.
|
|---|
| 22 | </p>
|
|---|
| 23 | <p>
|
|---|
| 24 | Depending on the specific hardware and host OS, this effect can
|
|---|
| 25 | very significantly skew the CPU usage reported by the OS. The
|
|---|
| 26 | reported CPU usage can be several times higher than what it
|
|---|
| 27 | would have been had the CPU been running at full speed. The
|
|---|
| 28 | effect can be observed both on the host OS and in a guest OS.
|
|---|
| 29 | </p>
|
|---|
| 30 | </body>
|
|---|
| 31 |
|
|---|
| 32 | </topic>
|
|---|
