1NUMACTL(8) Linux Administrator's Manual NUMACTL(8)
2
6 numactl - Control NUMA policy for processes or shared memory
7
9 numactl [ --all ] [ --balancing ] [ --interleave nodes ] [ --preferred
10 node ] [ --preferred-many nodes ] [ --membind nodes ] [ --cpunodebind
11 nodes ] [ --physcpubind cpus ] [ --localalloc ] [--] command {arguments
12 ...}
13 numactl --show
14 numactl --hardware
15 numactl [ --huge ] [ --offset offset ] [ --shmmode shmmode ] [ --length
16 length ] [ --strict ]
17 [ --shmid id ] --shm shmkeyfile | --file tmpfsfile
18 [ --touch ] [ --dump ] [ --dump-nodes ] memory policy
19
21 numactl runs processes with a specific NUMA scheduling or memory place‐
22 ment policy. The policy is set for command and inherited by all of its
23 children. In addition it can set persistent policy for shared memory
24 segments or files.
25 Use -- before command if using command options that could be confused
27 with numactl options.
28 nodes may be specified as N,N,N or N-N or N,N-N or N-N,N-N and so
30 forth. Relative nodes may be specified as +N,N,N or +N-N or +N,N-N
31 and so forth. The + indicates that the node numbers are relative to the
32 process' set of allowed nodes in its current cpuset. A !N-N notation
33 indicates the inverse of N-N, in other words all nodes except N-N. If
34 used with + notation, specify !+N-N. When same is specified the previ‐
35 ous nodemask specified on the command line is used. all means all
36 nodes in the current cpuset.
37 Instead of a number a node can also be:
39 netdev:DEV The node connected to network device DEV.
41 file:PATH The node the block device of PATH.
42 ip:HOST The node of the network device of HOST
43 block:PATH The node of block device PATH
44 pci:[seg:]bus:dev[:func] The node of a PCI device.
45 Note that block resolves the kernel block device names only for udev
47 names in /dev use file:
48 Policy settings are:
50 --all, -a
52 Unset default cpuset awareness, so user can use all possible
53 CPUs/nodes for following policy settings.
54 --interleave=nodes, -i nodes
56 Set a memory interleave policy. Memory will be allocated using
57 round robin on nodes. When memory cannot be allocated on the
58 current interleave target fall back to other nodes. Multiple
59 nodes may be specified on --interleave, --membind and --cpunode‐
60 bind.
61 --membind=nodes, -m nodes
63 Only allocate memory from nodes. Allocation will fail when
64 there is not enough memory available on these nodes. nodes may
65 be specified as noted above.
66 --cpunodebind=nodes, -N nodes
68 Only execute command on the CPUs of nodes. Note that nodes may
69 consist of multiple CPUs. nodes may be specified as noted
70 above.
71 --physcpubind=cpus, -C cpus
73 Only execute process on cpus. This accepts cpu numbers as shown
74 in the processor fields of /proc/cpuinfo, or relative cpus as in
75 relative to the current cpuset. You may specify "all", which
76 means all cpus in the current cpuset. Physical cpus may be
77 specified as N,N,N or N-N or N,N-N or N-N,N-N and so forth.
78 Relative cpus may be specified as +N,N,N or +N-N or +N,N-N and
79 so forth. The + indicates that the cpu numbers are relative to
80 the process' set of allowed cpus in its current cpuset. A !N-N
81 notation indicates the inverse of N-N, in other words all cpus
82 except N-N. If used with + notation, specify !+N-N.
83 --localalloc, -l
85 Try to allocate on the current node of the process, but if mem‐
86 ory cannot be allocated there fall back to other nodes.
87 --preferred=node
89 Preferably allocate memory on node, but if memory cannot be al‐
90 located there fall back to other nodes. This option takes only
91 a single node number. Relative notation may be used.
92 --balancing, -b
94 Enable Linux kernel NUMA balancing for the process if it is sup‐
95 ported by kernel. This should only be used with --membind, -m
96 only, otherwise ignored.
97 --preferred-many=node
99 Preferably allocate memory on nodes, but if memory cannot be al‐
100 located there fall back to other nodes. This option takes a
101 mask of preferred nodes where the closest node to local is con‐
102 sidered most preferred. Relative notation may be used.
103 --show, -s
105 Show NUMA policy settings of the current process.
106 --hardware, -H
108 Show inventory of available nodes on the system.
109 Numactl can set up policy for a SYSV shared memory segment or a file in
111 shmfs/hugetlbfs.
112 This policy is persistent and will be used by all mappings from that
114 shared memory. The order of options matters here. The specification
115 must at least include either of --shm, --shmid, --file to specify the
116 shared memory segment or file and a memory policy like described above
117 ( --interleave, --localalloc, --preferred, --preferred-many, --membind
118 ).
119 --huge
121 When creating a SYSV shared memory segment use huge pages. Only valid
122 before --shmid or --shm
123 --offset
125 Specify offset into the shared memory segment. Default 0. Valid units
126 are m (for MB), g (for GB), k (for KB), otherwise it specifies bytes.
127 --strict
129 Give an error when a page in the policied area in the shared memory
130 segment already was faulted in with a conflicting policy. Default is to
131 silently ignore this.
132 --shmmode shmmode
134 Only valid before --shmid or --shm When creating a shared memory seg‐
135 ment set it to numeric mode shmmode.
136 --length length
138 Apply policy to length range in the shared memory segment or make the
139 segment length long Default is to use the remaining length Required
140 when a shared memory segment is created and specifies the length of the
141 new segment then. Valid units are m (for MB), g (for GB), k (for KB),
142 otherwise it specifies bytes.
143 --shmid id
145 Create or use a shared memory segment with numeric ID id
146 --shm shmkeyfile
148 Create or use a shared memory segment, with the ID generated using
149 ftok(3) from shmkeyfile
150 --file tmpfsfile
152 Set policy for a file in tmpfs or hugetlbfs
153 --touch
155 Touch pages to enforce policy early. Default is to not touch them, the
156 policy is applied when an applications maps and accesses a page.
157 --dump
159 Dump policy in the specified range.
160 --dump-nodes
162 Dump all nodes of the specific range (very verbose!)
163 Valid node specifiers
165 all All nodes
167 number Node number
168 number1{,number2} Node number1 and Node number2
169 number1-number2 Nodes from number1 to number2
170 ! nodes Invert selection of the following specification.
171
173 numactl --physcpubind=+0-4,8-12 myapplic arguments Run myapplic on cpus
174 0-4 and 8-12 of the current cpuset.
175 numactl --interleave=all bigdatabase arguments Run big database with
177 its memory interleaved on all CPUs.
178 numactl --cpunodebind=0 --membind=0,1 process Run process on node 0
180 with memory allocated on node 0 and 1.
181 numactl --cpunodebind=0 --membind=0,1 -- process -l Run process as
183 above, but with an option (-l) that would be confused with a numactl
184 option.
185 numactl --cpunodebind=0 --balancing --membind=0,1 process Run process
187 on node 0 with memory allocated on node 0 and 1. Optimize the page
188 placement with Linux kernel NUMA balancing mechanism if possible.
189 numactl --cpunodebind=netdev:eth0 --membind=netdev:eth0 network-server
191 Run network-server on the node of network device eth0 with its memory
192 also in the same node.
193 numactl --preferred=1 numactl --show Set preferred node 1 and show the
195 resulting state.
196 numactl --preferred-many=0x3 numactl --show Set preferred nodes 1 and
198 2, and show the resulting state.
199 numactl --interleave=all --shm /tmp/shmkey Interleave all of the sysv
201 shared memory region specified by /tmp/shmkey over all nodes.
202 Place a tmpfs file on 2 nodes:
204 numactl --membind=2 dd if=/dev/zero of=/dev/shm/A bs=1M count=1024
205 numactl --membind=3 dd if=/dev/zero of=/dev/shm/A seek=1024 bs=1M
206 count=1024
207 numactl --localalloc /dev/shm/file Reset the policy for the shared mem‐
210 ory file file to the default localalloc policy.
211
213 Requires a NUMA policy aware kernel.
214 Command is not executed using a shell. If you want to use shell
216 metacharacters in the child use sh -c as wrapper.
217 Setting policy for a hugetlbfs file does currently not work because it
219 cannot be extended by truncate.
220 Shared memory segments larger than numactl's address space cannot be
222 completely policied. This could be a problem on 32bit architectures.
223 Changing it piece by piece may work.
224 The old --cpubind which accepts node numbers, not cpu numbers, is dep‐
226 recated and replaced with the new --cpunodebind and --physcpubind op‐
227 tions.
228
231 /proc/cpuinfo for the listing of active CPUs. See proc(5) for details.
232 /sys/devices/system/node/node*/numastat for NUMA memory hit statistics.
234
237 Copyright 2002,2004 Andi Kleen, SuSE Labs. numactl and the demo pro‐
238 grams are under the GNU General Public License, v.2
239
242 set_mempolicy(2) , get_mempolicy(2) , mbind(2) , sched_setaffinity(2) ,
243 sched_getaffinity(2) , proc(5) , ftok(3) , shmat(2) , migratepages(8)
244SuSE Labs Mar 2004 NUMACTL(8)