Chapter 6. Overcommitting with KVM
The KVM hypervisor automatically overcommits CPUs and memory. This means that more virtualized CPUs and memory can be allocated to virtual machines than there are physical resources on the system. This is possible because most processes do not access 100% of their allocated resources all the time.
As a result, under-utilized virtualized servers or desktops can run on fewer hosts, which saves a number of system resources, with the net effect of less power, cooling, and investment in server hardware.
6.1. Overcommitting Memory
Guest virtual machines running on a KVM hypervisor do not have dedicated blocks of physical RAM assigned to them. Instead, each guest virtual machine functions as a Linux process where the host physical machine's Linux kernel allocates memory only when requested. In addition the host's memory manager can move the guest virtual machine's memory between its own physical memory and swap space.
Overcommitting requires allotting sufficient swap space on the host physical machine to accommodate all guest virtual machines as well as enough memory for the host physical machine's processes. As a basic rule, the host physical machine's operating system requires a maximum of 4GB of memory along with a minimum of 4GB of swap space. For advanced instructions on determining an appropriate size for the swap partition, see the Red Hat KnowledgeBase.
Overcommitting is not an ideal solution for general memory issues. The recommended methods to deal with memory shortage are to allocate less memory per guest, add more physical memory to the host, or utilize swap space.
A virtual machine will run slower if it is swapped frequently. In addition, overcommitting can cause the system to run out of memory (OOM), which may lead to the Linux kernel shutting down important system processes. If you decide to overcommit memory, ensure sufficient testing is performed. Contact Red Hat support for assistance with overcommitting.
Overcommitting does not work with all guest virtual machines, but has been found to work in a desktop virtualization setup with minimal intensive usage or running several identical guest virtual machines with kernel same-page merging (KSM).
For more information on KSM and overcommitting, refer to Chapter 7, KSM.
When device assignment is in use, all virtual machine memory must be statically pre-allocated to enable direct memory access (DMA) with the assigned device. Memory overcommit is therefore not supported with device assignment.