Hi All,


I have enabled SNMP (NET-SNMP) agent on Red Hat V6.2.

We used to use ssCpuIdle as it is already presented in overall CPU utlisation in %.

However, as known, ssCpuIdle is already deprecated and replaced by ssCpuRawIdle, which is an integer counter.



We have been using the “monitor” method, with each monitor statement, associated with one OID parameter as below example:

monitor –r 15 “High CPU Utilization" ssCpuIdle< 25

Tried to make use of ssCpuRawIdle in a monitor statement but was unable to find much examples for reference.



Kindly need your expertise on the following:

- Can ssCpuRawIdle be used in a monitor statement?

- How do I determine (the integer output of ssCpuRawIdle =How many Overall CPU idle (%)?

- Are there other means (eg. OID parameters) to set and monitor threshold for SNMP trap?


Please help me.
Thanks.

Hi Sy,

Sure, if it's an snmp counter, it can be used in a trap.

The Manual from UC Davis suggests
On a multi-processor system, the 'ssCpuRaw*'
counters are cumulative over all CPUs, so their
sum will typically be N*100 (for N processors).

SNMP trap should work, if you are simply concerned about monitoring CPU idle and triggering an alert/event when it crosses a pre-determined threshold I'd stick to trying to get that set up. However...

There are a lot of things to consider when you look at something like this, particularly are we talking about instances or metal? If these are instances I'd look at more closely monitoring the compute profile of my instances at the environmental level (QEMU-KVM/VmWare/Xen etc), since that's where it matters.

Hi, dijetlo

Thanks a lot for the reply.

Please see further clarification required below:

I understand that it can be used in a trap , but unlikely as a parameter alone in a "monitor" statment.
As I have taken reference from this following link: http://serverfault.com/questions/441...-ticks-in-snmp, seems to indicate that it does require more than 1 OID paramater to compute the idle / overall CPU utilisation (%)

Hence, from what I understand from the explanation in the link, assuming that the number of ticks is 100, then a computation is required as follow:
CPU idle (%) = [Sum(ssCpuRawUser, ssCpuRawNice, ssCpuRawSystem) / Sum(ssCpuRawUser, ssCpuRawNice, ssCpuRawSystem, ssCpuRawIdle)] * 100


I need to confirm whether my understanding is correct.
1) Whether ssCpuRawIdle can be used alone in a "monitor" statement, if yes, I really need help with some "monitor" examples OR
2) I will need to resort to Scripting to compute the above formula.


Kindly, all, please help me.
Thanks.

Hmmm... I see what you're saying, the MIB is just a counter, the integer it corresponds to only has value if we're talking about the integers delta over time.
The answer would seem to be to pop the trap every so many seconds and let the snmp monitor carry the overhead of the calculations on the %'s and alert you if a boundry is breached.

So yeah... scripting? There'll be some.

I have tried to come up with a Perl script to compute and return a value.

However, it crosses my mind on how I can integrate it with NET-SNMP;
1) Push the value to the SNMP agent module
2) Trigger an SNMP trap based on a configured threshold (where should I configure this?)
3) SNMP trap is recognised by the SNMP monitoring application

Can anyone please enlighten me?
Are there also really no other NET-SNMP oid for CPU utilisation other than a) ssCpu* and b) ssCpuRaw*?

I am also looking for Professional Services to complete this implementation.
If anyone has others in mind, please let me know as well.

Thanks.