Memory overcommit (or overcommitment) is a hypervisor feature that allows a virtual machine (VM) to use more memory space than the physical host has available. For example, virtualization platforms like VMware ESX allow a host server with 2 GB of physical memory to run four guest machines, each with 1 GB of memory space allocated.
The idea of memory overcommit may seem dangerous, because a computer will crash if physical memory is exhausted. In actual practice, however, overcommitment of server computing resources is harmless -- most VMs use only a small portion of the physical memory that is allocated to them. For the previous example, a guest machine with 1 GB of physical memory allocated to it might only need 300-400 MB, leaving 600-700 MB of allocated space unused. If all four example machines use 300 MB, the physical server will have 800 MB of its original 2 GB left over.
Still, some VMs may need all (or even more) of the memory that they have been allocated, while some VMs may need considerably less. Hypervisors such as ESX can identify idle memory and dynamically reallocate unused memory from some VMs to others that need more memory. If none of the current guest machines need additional memory, any idle physical memory can be used to host additional guest machines if necessary.