A question that occurred to me due to some stuff I've been working on. If we assign a PCI passthrough device to a VM (let's say a GPU) and the PCIe slot for that device is tied to a specific processor...will the scheduler attempt to start the VM on that processor's home NUMA node and run the vCPUs on that CPU package? I've looked at esxtop output on a 6.0 U3 host and the VMs did not appear to follow that behavior. Does anyone know the default behavior here or have a link to relevant documentation?vmMarkA
↧