Monitoring virtual cpu capacity

[pabloadmin]

Hi, we are using today many types of virtualization servers, VMware,
IBM Lpars, Xen and others. The physical hardware could be heterogeneous,
i mean, we use for example: power processors, x86 or risc.
Suppose we are using a VMware cluster, it can automaticaly move the virtual
machine an other fisical hardware, on any problem.
If i was monitoring the cpu use, it can be change, suppose we are use diferent hardware. If i was consuming 20% of cpu and then, on the change, it was consuming 80% and i only see the cpu graph, i has a problem.
Suppose this case, we are running a lpar partition on IBM system, we can change the cpu allocation automticaly. If i see only the graph and the cpu use was 30% and the next time was 90%, i have a problem.
This case es very common situation, when the users are out and we calculate de interest rate for example.
Is It possible get a metric with the system capacity?, I mean, amount of floating points operations or some thing similar to a benchmark. Any thing to determine when the system was changed.


thank's

[pabloadmin]

Check this for details on IBM systems:

http://www.ibm.com/developerworks/aix/library/au-aix5_cpu/

[dmarker]

I've never really liked dealing with CPU usage as a percentage. Once I started working on AIX I've become fond of the "Physcial CPU consumed" metric.

Try adding these to zabbix_agentd.conf on your LPAR's:

UserParameter=cpu.online,/usr/bin/lparstat -i |grep "Online Virtual" |awk '{print $5}'
UserParameter=cpu.pc, if [ `vmstat |grep -c ' pc '` == 1 ]; then vmstat 1 1|tail -1|awk '{print $(NF-1) }' ;else print "Error";fi

If you graph those together, you get to see your CPU usage scaled against the number of CPU's in the system. As you add and remove CPU's from a system the upper usage limit (and possibly the usage) will be properly reflected in the graph without additional effort.

[pabloadmin]

I will try, thanks