What does %CPU mean in the output of "top" command

[Hanry Zhou]

one of process xyz in the result of "top" command is showing 60%, but the the overall cpu utilization in Glace is only showing 34%.

I don't feel that %CPU really represent the cpu persontage this particular process is taking from all CPU's on the server. But, what does this represent?

Also, what value of the normal load should be on 8 cpu's server?

Thanks,

[Steven E. Protter · Accepted solution]

Shalom,

Two utilities, two different calculation methods. People could argue all day about which is better and will do so, especially if you reward points for it.

CPU % is bases on sampling and demand for CPU cycles. It changes millisecond to millisecond. Also note that taking performance data CHANGES system performance.

Try doing it over time:
http://www.hpux.ws/?p=6

Second question: Load factor, there is no such thing as a normal level. You can have a load factor of 40 and perfectly normal system response if there are a lot of processes waiting in line for CPU.

The answer there is it depends. I generally start getting excited when I pass 10, but I've seen many, many many situations with higher numbers and no problems at all.

SEP

[Laurent Menase]

glance measure is based on delta time between context changes, so it is precise

top measure is a sample made on every tic.
basically it count 1 in the 3 category at tic time: IDLE SYSTEM USER
So you can have 1 process which appears to use 100% of the cpu when glance say about 0%
it just means that that thread is running at tic time. and only at that time.

You can also have a thread holding about 100% of CPU but top indicate 0%.

so use glance for precise measures.

[Ganesan R]

Hi,

Any also top shows the usage of independant cpu utilisation whereas glance shows the overall utilisation.

If you have multiple cpus, the process which shows 60% is on one particular cpu not the total cpu utilisation.