Is it possible to explicitly specify a value for the static metrics such
as cpu_num?
We have a number of nodes in our cluster which have hyperthreading
enabled, where each physical CPU appears as two CPUs to the OS. Our
applications are not threaded, so we get little or no benefit from
having the hyperthreading enabled (yes, we could disable it in the BIOS,
but I would like to leave it enabled "just in case"). Further, we have
CPUs (in our head or I/O nodes, for instance) which will never be used
for computation, but we would like to continue monitoring the nodes for
network, load average, etc.
Regardless, our cluster utilization shows 50%, even when actual cluster
performance peaks near 100%. Our management finds this confusing, and
I've had to explain it on a number of occasions. I would like to be
able to adjust the number of CPUs reported per node to a more reasonable
value -- the actual number of processors installed in each node which is
used for computation.
In my configuration file, I currently have:
collection_group {
collect_once = yes
time_threshold = 1200
metric {
name = "cpu_num"
}
}
Because there are two physical processors with hyperthreading, it shows
as 4 CPUs. I would like to do something like this:
collection_group {
collect_once = yes
time_threshold = 1200
metric {
name = "cpu_num"
value = 2
}
}
Is this possible?
