Many people in computing, we seem fixated on "percentages" when talking about CPU use - We need to fix that.
When a computer has an absolutely fixed number of CPUs on the time scale of a few years the percentages of the box makes a lot of sense.
But in a virtualised world we will have to change.
For 10 years, we can do Dynamic CPU changes to a LPAR (Virtual Machine).
Yesterdays 40% used is today's 80% because the admin guy added a couple of CPUs over night.
If we look at the percentages and see a massive jump and hit the panic button to escalate the issue.
If we look at the Physical CPU use then we see no change at all and relax.
This might happen up to a couple of times a day - then our CPU percentages have sudden and drastic changed that are nothing to do with the application or RDBMS use of the CPUs!
For the last 5 years, we can change "the size of the CPU bucket" in milliseconds with Shared CPUs under POWER Hypervisor control.
So we have 5% of 28 CPUs, the 60% of 8 CPUs, then 20% of 11 CPUs and 60% of 17 CPUs.
And that was just the last minute!!! Is the CPU use going down or up ?
If we switch to Physical CPU use it is completely obvious and a nice steady graph.
But we need to consider groups of Virtual Machine or the whole machines worth.
Percentages are then meaningless until we know the "percentage of what" and can do all the calculations instantaneously.
And no one can? Its like doing 100 calculations on percentages or have 100 physical CPU use numbers to scan through to find which Virtual Machine is the heavy CPU user.
While it is true that percentages in Capped Virtual Machines are reported OK against Entitlement - this is a special case.
I don't see installed machines where all Virtual Machines are Capped.
The reason is it is both expensive and a waste of CPU resources - as unused CPUs can't be harvested and used else where.
We need to monitor physical CPU use so we can
compare Virtual Machines and
add them up to the whole machine number..
CPU Percentages are "old school", complicated to work with and often misleading.