An administrator enables VM High Availability with resource reservation (Guarantee mode) on a four-node AHV cluster whose containers are all RF2. Before adding more VMs, they review cluster capacity. What will the reservation most likely change?

AHV High Availability has two modes. Best effort restarts VMs from a failed host only if free resources happen to exist. Guarantee mode, also called HA with resource reservation, sets aside enough memory to restart every VM from a failed host: one host's worth when all containers are RF2, or two hosts' worth when any container is RF3. The reservation is spread across hosts as segments, so each host gives up a share. Because that memory is held back, it no longer counts as available for new VMs, and Prism will refuse power-on requests that would consume the failover headroom. On a four-node RF2 cluster you should expect usable memory for VMs to drop by roughly a quarter of the cluster's total. The reservation does not stop HA; it is the thing that makes HA restarts certain. It doesn't add capacity either; it earmarks existing capacity. And it is primarily about memory, because AHV does not oversubscribe memory by default while vCPUs can be oversubscribed, so memory is what decides whether a failed host's VMs can restart. Exam caveat: AHV has no per-VM memory reservation setting; failover headroom comes from HA Guarantee mode, and enabling memory overcommit on VMs changes how much real memory they consume. Operational check: compare total VM memory with cluster memory minus the HA reservation in Prism before scheduling a rolling upgrade.