1NUMACTL(8) Linux Administrator's Manual NUMACTL(8)
2
6 numactl - Control NUMA policy for processes or shared memory
7
9 numactl [ --all ] [ --interleave nodes ] [ --preferred node ] [ --mem‐
10 bind nodes ] [ --cpunodebind nodes ] [ --physcpubind cpus ] [
11 --localalloc ] [--] command {arguments ...}
12 numactl --show
13 numactl --hardware
14 numactl [ --huge ] [ --offset offset ] [ --shmmode shmmode ] [ --length
15 length ] [ --strict ]
16 [ --shmid id ] --shm shmkeyfile | --file tmpfsfile
17 [ --touch ] [ --dump ] [ --dump-nodes ] memory policy
18
20 numactl runs processes with a specific NUMA scheduling or memory place‐
21 ment policy. The policy is set for command and inherited by all of its
22 children. In addition it can set persistent policy for shared memory
23 segments or files.
24 Use -- before command if using command options that could be confused
26 with numactl options.
27 nodes may be specified as N,N,N or N-N or N,N-N or N-N,N-N and so
29 forth. Relative nodes may be specifed as +N,N,N or +N-N or +N,N-N and
30 so forth. The + indicates that the node numbers are relative to the
31 process' set of allowed nodes in its current cpuset. A !N-N notation
32 indicates the inverse of N-N, in other words all nodes except N-N. If
33 used with + notation, specify !+N-N. When same is specified the previ‐
34 ous nodemask specified on the command line is used. all means all
35 nodes in the current cpuset.
36 Instead of a number a node can also be:
38 netdev:DEV The node connected to network device DEV.
40 file:PATH The node the block device of PATH.
41 ip:HOST The node of the network device of HOST
42 block:PATH The node of block device PATH
43 pci:[seg:]bus:dev[:func] The node of a PCI device.
44 Note that block resolves the kernel block device names only for udev
46 names in /dev use file:
47 Policy settings are:
49 --all, -a
51 Unset default cpuset awareness, so user can use all possible
52 CPUs/nodes for following policy settings.
53 --interleave=nodes, -i nodes
55 Set a memory interleave policy. Memory will be allocated using
56 round robin on nodes. When memory cannot be allocated on the
57 current interleave target fall back to other nodes. Multiple
58 nodes may be specified on --interleave, --membind and --cpunode‐
59 bind.
60 --membind=nodes, -m nodes
62 Only allocate memory from nodes. Allocation will fail when
63 there is not enough memory available on these nodes. nodes may
64 be specified as noted above.
65 --cpunodebind=nodes, -N nodes
67 Only execute command on the CPUs of nodes. Note that nodes may
68 consist of multiple CPUs. nodes may be specified as noted
69 above.
70 --physcpubind=cpus, -C cpus
72 Only execute process on cpus. This accepts cpu numbers as shown
73 in the processor fields of /proc/cpuinfo, or relative cpus as in
74 relative to the current cpuset. You may specify "all", which
75 means all cpus in the current cpuset. Physical cpus may be
76 specified as N,N,N or N-N or N,N-N or N-N,N-N and so forth.
77 Relative cpus may be specifed as +N,N,N or +N-N or +N,N-N and
78 so forth. The + indicates that the cpu numbers are relative to
79 the process' set of allowed cpus in its current cpuset. A !N-N
80 notation indicates the inverse of N-N, in other words all cpus
81 except N-N. If used with + notation, specify !+N-N.
82 --localalloc, -l
84 Always allocate on the current node.
85 --preferred=node
87 Preferably allocate memory on node, but if memory cannot be
88 allocated there fall back to other nodes. This option takes
89 only a single node number. Relative notation may be used.
90 --show, -s
92 Show NUMA policy settings of the current process.
93 --hardware, -H
95 Show inventory of available nodes on the system.
96 Numactl can set up policy for a SYSV shared memory segment or a file in
98 shmfs/hugetlbfs.
99 This policy is persistent and will be used by all mappings from that
101 shared memory. The order of options matters here. The specification
102 must at least include either of --shm, --shmid, --file to specify the
103 shared memory segment or file and a memory policy like described above
104 ( --interleave, --localalloc, --preferred, --membind ).
105 --huge
107 When creating a SYSV shared memory segment use huge pages. Only valid
108 before --shmid or --shm
109 --offset
111 Specify offset into the shared memory segment. Default 0. Valid units
112 are m (for MB), g (for GB), k (for KB), otherwise it specifies bytes.
113 --strict
115 Give an error when a page in the policied area in the shared memory
116 segment already was faulted in with a conflicting policy. Default is to
117 silently ignore this.
118 --shmmode shmmode
120 Only valid before --shmid or --shm When creating a shared memory seg‐
121 ment set it to numeric mode shmmode.
122 --length length
124 Apply policy to length range in the shared memory segment or make the
125 segment length long Default is to use the remaining length Required
126 when a shared memory segment is created and specifies the length of the
127 new segment then. Valid units are m (for MB), g (for GB), k (for KB),
128 otherwise it specifies bytes.
129 --shmid id
131 Create or use an shared memory segment with numeric ID id
132 --shm shmkeyfile
134 Create or use an shared memory segment, with the ID generated using
135 ftok(3) from shmkeyfile
136 --file tmpfsfile
138 Set policy for a file in tmpfs or hugetlbfs
139 --touch
141 Touch pages to enforce policy early. Default is to not touch them, the
142 policy is applied when an applications maps and accesses a page.
143 --dump
145 Dump policy in the specified range.
146 --dump-nodes
148 Dump all nodes of the specific range (very verbose!)
149 Valid node specifiers
151 all All nodes
153 number Node number
154 number1{,number2} Node number1 and Node number2
155 number1-number2 Nodes from number1 to number2
156 ! nodes Invert selection of the following specification.
157
159 numactl --physcpubind=+0-4,8-12 myapplic arguments Run myapplic on cpus
160 0-4 and 8-12 of the current cpuset.
161 numactl --interleave=all bigdatabase arguments Run big database with
163 its memory interleaved on all CPUs.
164 numactl --cpunodebind=0 --membind=0,1 process Run process on node 0
166 with memory allocated on node 0 and 1.
167 numactl --cpunodebind=0 --membind=0,1 -- process -l Run process as
169 above, but with an option (-l) that would be confused with a numactl
170 option.
171 numactl --cpunodebind=netdev:eth0 --membind=netdev:eth0 network-server
173 Run network-server on the node of network device eth0 with its memory
174 also in the same node.
175 numactl --preferred=1 numactl --show Set preferred node 1 and show the
177 resulting state.
178 numactl --interleave=all --shm /tmp/shmkey Interleave all of the sysv
180 shared memory region specified by /tmp/shmkey over all nodes.
181 Place a tmpfs file on 2 nodes:
183 numactl --membind=2 dd if=/dev/zero of=/dev/shm/A bs=1M count=1024
184 numactl --membind=3 dd if=/dev/zero of=/dev/shm/A seek=1024 bs=1M
185 count=1024
186 numactl --localalloc /dev/shm/file Reset the policy for the shared mem‐
189 ory file file to the default localalloc policy.
190
192 Requires an NUMA policy aware kernel.
193 Command is not executed using a shell. If you want to use shell
195 metacharacters in the child use sh -c as wrapper.
196 Setting policy for a hugetlbfs file does currently not work because it
198 cannot be extended by truncate.
199 Shared memory segments larger than numactl's address space cannot be
201 completely policied. This could be a problem on 32bit architectures.
202 Changing it piece by piece may work.
203 The old --cpubind which accepts node numbers, not cpu numbers, is dep‐
205 recated and replaced with the new --cpunodebind and --physcpubind
206 options.
207
210 /proc/cpuinfo for the listing of active CPUs. See proc(5) for details.
211 /sys/devices/system/node/node*/numastat for NUMA memory hit statistics.
213
216 Copyright 2002,2004 Andi Kleen, SuSE Labs. numactl and the demo pro‐
217 grams are under the GNU General Public License, v.2
218
221 set_mempolicy(2) , get_mempolicy(2) , mbind(2) , sched_setaffinity(2) ,
222 sched_getaffinity(2) , proc(5) , ftok(3) , shmat(2) , migratepages(8)
223SuSE Labs Mar 2004 NUMACTL(8)