numactl
runs processes with a specific NUMA scheduling or memory placement policy.
The policy is set for command and inherited by all of its children.
In addition it can set persistent policy for shared memory segments or files.
Policy settings are:
--interleave=nodes, -i nodes
Set an memory interleave policy. Memory will be allocated using round robin
on
nodes.
When memory cannot be allocated on the current interleave target fall back
to other nodes.
--membind=nodes, -m nodes
Only allocate memory from nodes. Allocation will fail when there
is not enough memory available on these nodes.
--cpunodebind=nodes, -N nodes
Only execute process on the CPUs of
nodes.
Note that nodes may consist of multiple CPUs.
--physcpubind=cpus, -C cpus
Only execute process on
cpus.
This accepts physical cpu numbers as shown in the
processor
fields of
/proc/cpuinfo.
--localalloc, -l
Do always local allocation on the current node.
--preferred=node
Preferably allocate memory on
node,
but if memory cannot be allocated there fall back to other nodes.
This option takes only a single node number.
--show, -s
Show NUMA policy settings of the current process.
--hardware, -H
Show inventory of available nodes on the system.
Numactl can set up policy for a SYSV shared memory segment or a file in shmfs/hugetlbfs.
This policy is persistent and will be used by
all mappings from that shared memory. The order of options matters here.
The specification must at least include either of
--shm, --shmid, --file
to specify the shared memory segment or file and a memory policy like described
above (
--interleave, --localalloc, --prefered, --membind
).
--huge
When creating a SYSV shared memory segment use huge pages.
Only valid before --shmid or --shm
--offset
Specify offset into the shared memory segment. Default 0.
Valid units are
m
(for MB),
g
(for GB),
k
(for KB),
otherwise it specifies bytes.
--strict
Give an error when a page in the policied area in the shared memory
segment already was faulted in with a conflicting policy. Default
is to silently ignore this.
--mode shmmode
Only valid before --shmid or --shm
When creating a shared memory segment set it to numeric mode
shmmode.
--length length
Apply policy to
length
range in the shared memory segment or make
the segment length long
Default is to use the remaining length
Required when a shared memory segment is created and specifies the length
of the new segment then. Valid units are
m
(for MB),
g
(for GB),
k
(for KB),
otherwise it specifies bytes.
--shmid id
Create or use an shared memory segment with numeric ID
id
--shm shmkeyfile
Create or use an shared memory segment, with the ID generated
using
ftok(3)
from shmkeyfile
--file tmpfsfile
Set policy for a file in tmpfs or hugetlbfs
--touch
Touch pages to enforce policy early. Default is to not touch them, the policy
is applied when an applications maps and accesses a page.
--dump
Dump policy in the specified range.
Valid node specifiers
all
All nodes
number
Node number
number1{,number2}
Node number1 and Node number2
number1-number2
Nodes from number1 to number2
! nodes
Invert selection of the following specification.
EXAMPLES
numactl --interleave=all bigdatabase arguments
Run big database with its memory interleaved on all CPUs.
numactl --cpubind=0--membind=0,1 process
Run process on node 0 with memory allocated on node 0 and 1.
numactl --preferred=1 numactl --show
Set preferred node 1 and show the resulting state.
numactl --interleave=all --shmkeyfile /tmp/shmkey
Interleave all of the sysv shared memory regiion specified by
/tmp/shmkey over all nodes.
numactl --offset=1G --length=1G --membind=1 --file /dev/shm/A --touch
Bind the second gigabyte in the tmpfs file /dev/shm/A to node 1.
numactl --localalloc /dev/shm/file
Reset the policy for the shared memory file
file
to the default localalloc policy.
NOTES
Requires an NUMA policy aware kernel.
Command is not executed using a shell. If you want to use shell metacharacters
in the child use sh -c as wrapper.
Setting policy for a hugetlbfs file does currently not work because
it cannot be extended by truncate.
Shared memory segments larger than numactl's address space cannot
be completely policied. This could be a problem on 32bit architectures.
Changing it piece by piece may work.
The old
--cpubind
which accepts node numbers, not cpu numbers, is deprecated
and replaced with the new
--cpunodebind
and
--physcpubind
options.
FILES
/proc/cpuinfo
for the listing of active CPUs. See
proc(5)
for details.
/sys/devices/system/node/node*/numastat
for NUMA memory hit statistics.
COPYRIGHT
Copyright 2002,2004 Andi Kleen, SuSE Labs.
numactl and the demo programs are under the GNU General Public License, v.2
