Linux load numbers

Many utilities, such as top in [procps]( display the percentages of time the cpu is busy doing things such as userland programs, system calls or just idle. This page describes the file /proc/stat and how programs interpret the numbers they find.

I am the [Debian]( maintainer for procps which contains top. Often I get bug reports about those numbers that appear at the top of top (called the summary area) so hopefully it will
help Debian users understand it too.

##The /proc/stat file
The file /proc/stat file is where the cpu numbers come from. As I am typing this, my single Athlon cpu computer running Linux 2.6.15 had the first two lines of the file looking like:

$ grep ^cpu /proc/stat
cpu  217174 10002 105629 7692822 90422 6491 22673 0
cpu0 217174 10002 105629 7692822 90422 6491 22673 0

The first thing you can see is I have 1 cpu, as there is only the aggregate line (starting with cpu) and then one individual cpu line (showing cpu0). Each field is describing how much time the cpu is been in various states, the values are in jiffies (more about them later). From left to right, the values are:

* Userland – running normal programs
* Nice – running niced programs
* System – running processes at the system level, eg the kernel
* Idle – CPU is doing nothing (running idle task)
* IOwait – CPU is waiting for IO to come back
* irq – servicing a hardware interrupt
* softirq – servicing a software interrupt
* Steal – To do with virtual machines, this cpu is waiting for the others

Quite often the kernel doesn’t count time in seconds, but counts them in a unit called jiffies. There is a concept of a value called Hz or Hertz which is the number of jiffies in a second. Happily for us, we’re only
looking at percentages, so it doesn’t really matter.