vHT Full CPU Reservation

You can use a formula to calculate the full CPU reservation for vHT.

For a low latency VM without vHT, each vCPU of the VM has exclusively affinity to a thread of the physical core. For hosts with Hyperthreading activated, the partner hyperthread has exclusive affinity to an idle world. Each vCPU of a low latency VM is allocated a dedicated physical core.

The CPU reservation for low latency VM is computed as:

Low latency VM (without vHT)
CPU minimum reservation = numVcpus * cpuFrequency

However, when vHT is activated for a VM, each hypertwin of the physical core is shared between multiple vCPUs of the VM where each hypertwin has exclusive affinity to a vCPU of the VM. This means a core with numSMT physical hyperthreads is shared by numSMT virtual threads. In this case, the CPU reservation requirement would be computed as:

Low latency VM (with vHT)
CPU minimum reservation = (numVcpus / numSMT) * cpuFrequency

Table 1. Example of booting a low latency VM with 20 vCPUs on a host with CPU frequency as 2 GHz

	numSMT = 1(without vHT)	numSMT = 2(with vHT)
numVcpus	20	20
number of Physical cores	20	10 (each core shared by 2 vCPUs)
Required minimum CPU reservation	20 * 2.0 GHz = 40 GHz	(20/2) * 2.0 GHz = 20 GHz