When starting a vmi with guaranteed QOS, the resulting virt-launcher pod is very slow. After some analysis, it shows that the cgroups are configured as the following: 
```
/<path-to-pod>/<our-container>/cpu.max (1000 100000)
/<path-to-pod>/<our-container>/container/cpu.max (max 100000)
```
Which means that the cpu is throttled.

The same pod with 4.19 the cgroups:
```
/<path-to-pod>/<our-container>/cpu.max (200000 100000)
/<path-to-pod>/<our-container>/container/cpu.max (200000 100000)
```

4.20    

Everytime

    1. Start a vm with dedicated cpu
    2. Look at the cgroups in the node where the pod has been scheduled
    3. Try to console the vmi
    

The vmi is booting very very slow

The vmi booting at normal speed

U/S cri-o issue: https://github.com/cri-o/cri-o/issues/9251


We also noticed a general slowness in the runtime of the CI with provider 1.33. In particular, with 1.32 provider the testsuite takes ~1h30m; against the 4h20m with provider 1.33.
My personal suspect is that, even with non guuaranteed QOS, there is something wrong.