everything I want gets picked up from the dynamo process.
Although for some reason the CPU Utilization isn't being presented and recorded.
Any ideas?
Thank you
EDIT:
SYSTEM - Linux x86_64, Virtual Machine, running atop ESX4.0 on an IBM HS21 Blade
Asked
Active
Viewed 369 times
1
-
I'm assuming Linux, is that correct? – James Polley Jan 08 '10 at 06:31
-
Just to clarify, everything works perfectly except the CPU Utilization? I believe that's what you said but I want to make sure I understand. – Joshua Nurczyk Jan 08 '10 at 17:00
-
yes - my mistake. Linux it is. x86_64. It's a VM to boot. – romant Jan 09 '10 at 22:22
1 Answers
1
Whether your tools works or not CPU measurement from within a VM is utterly pointless, any data you get is flawed. This is because the OS and/or tool assumes it has full control of the (actually virtual) CPU so measurement is based on load over time and time isn't linear in VM-land due to time-slicing.
Measure from the hypervisor, it's the only way to know what's going on.
Chopper3
- 100,240
- 9
- 106
- 238
-
You're absolutely right - if there is contention for the CPU. In my case this is the only virtual machine running on the host apart from the ESX console itself. Understandably there is a slight hit in the guarantee and scheduling that is incurred from ESX, but am not interested in those variations. Am after the cost per io operation. – romant Jan 17 '10 at 13:40
-
Host utilisation is irrelevant - VM CPU measurement is **ALWAYS** wrong, in all circumstances - it's actually a question on in the VCP 3 & 4 exams. – Chopper3 Jan 17 '10 at 13:55
-
Chopper, yes - its off the 100% ideal. This isn't the issue here, as the IO cost will still be the same, as whilst the VM is 'sleeping' in between the ESX-CPU scheduling, so is the IO within that machine. – romant Jan 18 '10 at 04:29