Hi,
I deploy a ESXi nested hypervisor and have have 19 VM's runing in the virtual ESX, half of them are 64 bit OS's.
I start to see some problem swith CPU competition by the VM's, where in esxtop the RDY field is between 10-40 to some machines, and clearly I have a problem.
I have a physical machine with two 6 core CPU's (X5690) and I also notide that when I define in the virtual ESX 20 core ( left some cores out for the VMK ) the perfomance of the CPU imporved a lot, I assume that is related to the fact that the Virtual ESX doesn't support Hyperthreading.
Now what I notice is that the virtual ESX report more than twice the MHZ that the physical machine for example when I see 16104 MHz ( Max capacity 41484 MHz ) in the physical ESX I see 35684 MHz in the Virtual ESX.
In the esxtop I see in the physical machine :
6:06:18pm up 32 days 6:01, 314 worlds, 1 VMs, 20 vCPUs; CPU load average: 0.50, 0.50, 0.45
PCPU USED(%): 20 19 21 16 18 16 19 14 18 16 18 19 26 23 21 21 18 27 24 23 24 17 15 26 AVG: 20
PCPU UTIL(%): 29 28 27 22 24 21 26 21 27 24 26 27 37 33 28 28 28 37 34 33 33 24 22 35 AVG: 28
CORE UTIL(%): 44 41 39 39 39 43 54 46 51 53 45 45 AVG: 45
And in the Virtual ESX :
9:59:21pm up 46 min, 442 worlds, 19 VMs, 28 vCPUs; CPU load average: 0.93, 0.63, 0.52
PCPU USED(%): 74 82 81 78 76 67 90 81 62 22 75 78 69 80 81 73 76 79 80 76 AVG: 74
PCPU UTIL(%): 75 82 83 78 76 71 91 84 64 24 76 77 70 80 82 75 76 80 81 77 AVG: 75
The problem that I see is that I run out of CPU in the virtual machine when in the physical one I still have plenty of cycles.
Any optimization in the scheduller that can improve the behaivour ? or any other thing that I can improve ?
Paulo
