A metric for overall system utilization

Jim Mauro

2011-06-01 13:39:22 UTC

Hi Kishore.

It depends on what you mean by "overall system utilization". Most people
mean CPU utilization, as opposed to utilization of the network or disk
subsystems.

CPU utilization is a combination of usr + sys, so the sum of those two
(or 100 - idle).

Many like to use load average as a "how busy" metric, which combines
CPU usr + sys and run queue wait times. Note, to my mind, load average
must be correlated with a workload/application performance metric
to be of real use. Actually, any system performance work requires
a measure of application performance, in conjunction with the system
metrics.

I use a combination of CPU utilization (vmstat, mpstat), load average,
run queue depth (vmstat) and CPU wait times (LAT thread microstate
as reported by prstat -Lm).

HTH
/jim

Post by Kishore Kumar Pusukuri
Hi,
Could you please provide me some metrics used for measuring overall system utilization (e.g. a NUMA machine running Solaris 10)?
Is it okay to use overall CPU utilization metric such as "usr" under mpstat(1) output?
Please let me know.
Thanks.
--
This message posted from opensolaris.org
_______________________________________________
perf-discuss mailing list