This might be a rookie question but if I'm currently on a 4:1 oversubscription of my CPU ressources does that mean that my processors performances are divided(splitted) into 4 lower perf. vCPU? Will a vm using 4 vCPU will just divide a pCPU into 4 to actually be as fast as the pCPU itself?
-
1This is quite a complex topic, but it really just boils down to what the CPU usage of your VMs are. If they're not using much CPU, you can do a 20:1 "oversubscription" without any notable performance impact. The key to you sorting this out is the same as all capacity planning topics: testing your actual workload, on actual hardware, with your proposed configuration. Then see if the performance is up to your requirements. – EEAA Sep 03 '14 at 02:59
1 Answers
If I understand your question correctly the answer is "no". However, it depends on your workload.
ESXi schedules your vCPUs to run on pCPUs. You have a vCPU/pCPU ratio of 4:1. If 25% of your VMs request 100% CPU time and the rest is running at 0MHz, the 25% will get the full CPU performance. (Minus hypervisor overhead, which is said to be 5 to 6% for very demanding workloads.)
If all your VMs request 25% CPU time your pCPUs will run at 100% but your VMs get the CPU they need.
If all of your VMs request 100% CPU time, the physically available CPU ressources will be divided between them. In that case, they will probably get roughly 1/4 of the performance of your physical cores.
This is a simplified explanation not taking into account reservations, limits, data locality and other stuff but I hope it helps.
- 1,612
- 9
- 13