vmovercommitmemory

In Linux, vm.overcommit_memory (often discussed as vmovercommitmemory) is a kernel parameter that controls how the system handles memory allocations that may exceed available physical RAM and swap. The setting is exposed through /proc/sys/vm/overcommit_memory and can be changed at runtime with sysctl or by writing to the proc file. It influences allocation decisions, the likelihood of allocation failures, and the activation of the out-of-memory (OOM) killer.

The parameter has three values. 0 enables a heuristic overcommit policy where the kernel decides whether to allow an allocation based on current memory usage and swap. This provides a balance between performance and safety, but can still lead to OOM conditions under pressure. 1 allows all allocations to succeed regardless of memory state, which can improve application availability but raises the risk of triggering OOM conditions when memory is actually exhausted. 2 disallows overcommit beyond a calculated limit, forcing the kernel to deny allocations that would exceed RAM plus swap by a configurable margin.

When vm.overcommit_memory is set to 2, the kernel uses vm.overcommit_ratio (a percentage) to determine how much memory may be allocated. The allowable overcommitment is roughly (RAM + SWAP) multiplied by overcommit_ratio/100. This mode reduces the chance of sudden OOM events but can cause allocations to fail with ENOMEM even if some free memory exists.

To view and set: cat /proc/sys/vm/overcommit_memory and echo 0 | sudo tee /proc/sys/vm/overcommit_memory. To make changes persistent, add vm.overcommit_memory=0 (or 1/2) and vm.overcommit_ratio=value to /etc/sysctl.conf or a file under /etc/sysctl.d/.