Hi everyone,
I have a concern regarding the relation between the % of CPU Ghz cycles defined in an allocation pool and the vCPU assignment and possible overcommitment.
Let's assume we are running latest stable versions of vCloud Director as well as the underlying vSphere environnement.
When it comes to propose virtual CPU ressources to customers, we ask them the number of vCPU's they need for their VM's.
In the pool (allocation type), 1 vCPU is defined at 2000Mhz. If the customers needs a total of 100 vCPU's for his VM's, a reservation of 200.000Mhz is defined in that pool.
Let's say that for the whole compute ressources we have a total of 316.800Mhz (144 cores * 2200Mhz).
Now, here comes the part which is not 100% clear for me :
In reality, VM's from that customer (with 100vCPU defined amongst them) are consuming maximum 20.000Mhz of computer power.
Question 1 : First of all, I guess the most logic (and transparent) way of working would be to ask our customers how many Ghz they are consuming on their current environment in order to size their cloud environnement properly ? If I was customer, I would not see the point of paying for 200Ghz when using only 20Ghz.
Question 2 : This allocation is taken into account ONLY under contention, right ? Which means that in case of a 20.000Mhz usage within a 200.000Mhz allocation, other VM's could use the remaining ressources regardless of the allocation model ?
Question 3 : about the CPU guarantee part : what does it mean exacly ? that under contention, the amount of clock cycles are given in priority to the VM's running in that pool ? what is the difference with CPU allocation ? What sense does it make to allocate ressources but not guarantee them ? or is it like some CPU thin provisionning ? like if the allocation works like a possible burst under non-contention but that the guarantee %age is kind of reserved ? Do I see this right ?
Question 4 : the difference between an allocation pool and a reservation pool is that the reservation pool guarantees 100% of CPU ressources ?
Question 5 : Let's say that we have 6 physical ESXi host's contributing to those 316.800Mhz. Those host's are splitted accross 2 datacenters. What happens if we loose 3 hosts ? The 200.000Mhz allocation pool exceeds the real computing power of the 3 remaining host's, will the concerned VM's HA failovered to the remaining 3 host's ?
Question 6 : What is the impact of vCPU overcommitment within an given pool regarding physical cores ? I guess CPU READY times who jumps up or is the impact less than before ? With 3 host I would only have 72 physical core's left when minimum 100vCPU's are alocated. But I guess it all depends of what those 100vCPU's are doing. If they work 100% performance will drop, if they work 10%, we won't see the difference. Is this right ?
Thank's for your help 