Hi,
I am trying to figure out accurate thread CPU utilization using the sched
provider. When I sample for 5 seconds and multiply by the number of CPU
threads, the reported seconds exceeds 5s X number of CPU threads.
sched:::on-cpu
/self->ts == 0/
{
self->ts = timestamp/nanosec;
}
sched:::off-cpu
/self->ts != 0/
{
@a[pid,tid] = sum((timestamp/nanosec) - self->ts);
self->ts = 0;
}
tick-5s {
trunc(@a,10);
printa(@a);
}
=====================
Here is the output
3-------------1----------3
1982----------165-------3
6039----------52571---3
12110---------6006-----6
1982----------25785---8
1982----------25930--8
12110---------5996-----9
6039----------52902----12
11359----------3006-----14
0-------------0----------4751
The total of the above exceeds 5X256 (T5440 with 256CPU threads). Any ideas?
Thanks
--
This message posted from opensolaris.org
_______________________________________________
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
