numactl
runs processes with a specific NUMA scheduling or memory placement policy
TLDR
Run command on node 0 with memory on nodes 0 and 1
SYNOPSIS
numactl [--interleave=nodes] [--membind=nodes] [--cpunodebind=nodes] [--physcpubind=cpus] [--] command [arguments]
DESCRIPTION
numactl runs processes with a specific NUMA (Non-Uniform Memory Access) scheduling or memory placement policy. The policy is set for the command and inherited by all its children. It can also manage policies for shared memory segments.
PARAMETERS
-a, --all
Disable cpuset awareness for broader CPU/node access-i, --interleave=nodes
Round-robin memory allocation across specified nodes-m, --membind=nodes
Restrict memory allocation to specified nodes only-N, --cpunodebind=nodes
Limit CPU execution to processors on specified nodes-C, --physcpubind=cpus
Bind process to specific physical CPUs-l, --localalloc
Prefer allocation on current node with fallback-p, --preferred=node
Prefer single node for allocation with fallback-P, --preferred-many=nodes
Prefer multiple nodes based on proximity-b, --balancing
Enable kernel NUMA balancing-s, --show
Display current NUMA policy settings-H, --hardware
Show available nodes and CPUs--huge
Use huge pages for SYSV shared memory--shm, --shmid
Specify shared memory segment--touch
Enforce policy immediately rather than on access
CAVEATS
NUMA policies only make sense on multi-node systems. Node numbering starts at 0. The special value "all" can be used to specify all nodes.
HISTORY
numactl is part of the numactl package for Linux NUMA systems. It provides user-space control over NUMA memory and CPU policies that complement kernel automatic NUMA balancing.
SEE ALSO
migratepages(8), numastat(8), set_mempolicy(2), mbind(2)