Measure ESXi memory overcommit and see which reclamation stage it triggers: page sharing, ballooning, compression or host swap.
ESXi reclaims memory in escalating stages. Page sharing is free and always on, ballooning asks the guest to give back idle pages, compression stores reclaimed pages at roughly 2:1 in a memory cache, and host swapping writes them to disk — the only stage that genuinely hurts. Overcommit above 1.0 is normal and safe as long as the summed active working set still fits in physical RAM. Overcommit alone is not a problem; a working set that exceeds physical memory is, because host swapping turns nanosecond memory access into millisecond disk access and every VM on the host feels it at once.
VMware Memory Overcommit
overcommit = VMs × (configured RAM + VMX overhead) ÷ (installed RAM × (1 − host overhead)); the stage is chosen from that ratio and from whether the active working set still fits.
overcommit = VMs × (configured RAM + VMX overhead) ÷ (installed RAM × (1 − host overhead)); the stage is chosen from that ratio and from whether the active working set still fits. ESXi reclaims memory in escalating stages. Page sharing is free and always on, ballooning asks the guest to give back idle pages, compression stores reclaimed pages at roughly 2:1 in a memory cache, and host swapping writes them to disk — the only stage that genuinely hurts. Overcommit above 1.0 is normal and safe as long as the summed active working set still fits in physical RAM.
Overcommit alone is not a problem; a working set that exceeds physical memory is, because host swapping turns nanosecond memory access into millisecond disk access and every VM on the host feels it at once.
This calculator takes 8 inputs: Hosts in cluster, Installed memory per host, ESXi host memory overhead, Powered-on VMs, Configured memory per VM, VMX overhead per VM, Active working set per VM, Page sharing and compression savings. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
Yes. A full reservation removes that VM's memory from the reclaimable pool entirely and also eliminates its swap file, so heavy use of reservations lowers the overcommit you can safely run across the rest of the cluster.