MRTG: ssCpuRawUser and System didn't report correct value

[Henry]

Dear folks,

I have a 4 CPU server (total 7 logical processors with HT) configured
with MRTG CPU monitoring and found that ssCpuRawUser and
ssCpuRawSystem didn't report correct usage values. I tried to stress
the server with loading, which sar reported 100% user load vs the
ssCPURawUser in MRTG reported drop in user usage ..

I also did the same experiment in a another single CPU system (2
logical processors), the report is fine for both MRTG and sar.

do anyone have idea on what is wrong??

Many thanks!

Henry

[Niels Baggesen]

Post by Henry
I have a 4 CPU server (total 7 logical processors with HT) configured
with MRTG CPU monitoring and found that ssCpuRawUser and
ssCpuRawSystem didn't report correct usage values. I tried to stress
the server with loading, which sar reported 100% user load vs the
ssCPURawUser in MRTG reported drop in user usage ..

net-snmp does not scale the counters returned from OS to the number of
CPUs, so with 7 cores you can expect utilization to reach 700% (except
on Solaris where the kernel scales the numbers - as far as I remember)

You problably have you measurements dropped because they exceed your
maximum value of 100%

/Niels

[Henry]

Thanks for your reply!

btw, any idea on the scaling formula? I think I could make the
correction in mrtg cfg file.

PS: my server should have 8 cores, which numbered from 0-7.

Post by Niels Baggesen
net-snmp does not scale the counters returned from OS to the number of
CPUs, so with 7 cores you can expect utilization to reach 700% (except
on Solaris where the kernel scales the numbers - as far as I remember)
You problably have you measurements dropped because they exceed your
maximum value of 100%
/Niels
--
The purpose of computing is insight, not numbers  --  R W Hamming

[Henry]

I found that I could get a correct 100% usage by dividing the
ssCpuRawUser + System utilization by no of cores (i.e. 8 for my
server). It is strange that the scaling is correct for a two cores
server but a 8 cores one..

Post by Henry
Thanks for your reply!
btw, any idea on the scaling formula? I think I could make the
correction in mrtg cfg file.
PS: my server should have 8 cores, which numbered from 0-7.

Post by Niels Baggesen
net-snmp does not scale the counters returned from OS to the number of
CPUs, so with 7 cores you can expect utilization to reach 700% (except
on Solaris where the kernel scales the numbers - as far as I remember)
You problably have you measurements dropped because they exceed your
maximum value of 100%
/Niels
--
The purpose of computing is insight, not numbers  --  R W Hamming

[Niels Baggesen]

Post by Henry
I found that I could get a correct 100% usage by dividing the
ssCpuRawUser + System utilization by no of cores (i.e. 8 for my
server). It is strange that the scaling is correct for a two cores
server but a 8 cores one..

How busy is the two core system? Does it ever keep more than one
core busy?

/Niels

[Henry]

The two core server didn't have application running. I could get 50%
usage with one looping script and 100% for two.

thanks and regards,

Henry

Post by Niels Baggesen
How busy is the two core system? Does it ever keep more than one
core busy?
/Niels
--
The purpose of computing is insight, not numbers  --  R W Hamming