Thanks for you reply.
As you say, 90% doesn't use admission control. For us, it's has no sens to use it because in case of needed (for example multiple host failure), we want to be able to run in degraded mode by using ressource pool. VM with high ressource pool (production vm) will be prioritize, and if necessary we will be able to power off test VM to enable power to Production VM.
We have several stretched clusters that are active active server on 2 datacenter. If a datacenter fails, HA will restart VM on the other site. So, to achieve this, we want to limit usable capacity to 50%. It was done under vrops by a dedicated policy. With that policy i was able to know how many VM/Time remains until 50%.
Now, the metrics is 100%, and the vm remaining value is not what i'm expected.