numa - overview of Non-Uniform Memory Architecture
DESCRIPTION
Non-Uniform Memory Access (NUMA) refers to multiprocessor systems
whose memory is divided into multiple memory nodes.
The access time of a memory node depends on
the relative locations of the accessing CPU and the accessed node.
(This contrasts with a symmetric multiprocessor system,
where the access time for all of the memory is the same for all CPUs.)
Normally, each CPU on a NUMA system has a local memory node whose
contents can be accessed faster than the memory in
the node local to another CPU
or the memory on a bus shared by all CPUs.
NUMA system calls
The Linux kernel implements the following NUMA-related system calls:
get_mempolicy(2),
mbind(2),
migrate_pages(2),
move_pages(2),
and
set_mempolicy(2).
However, applications should normally use the interface provided bu
libnuma;
see "Library Support" below.
/proc/[number]/numa_maps (since Linux 2.6.14)
This file displays information about a process's
NUMA memory policy and allocation.
Each line contains information about a memory range used by the process,
displaying---among other information---the effective memory policy for
that memory range and on which nodes the pages have been allocated.
numa_maps
is a read-only file.
When
/proc/<pid>/numa_maps
is read, the kernel will scan the virtual address space of the
process and report how memory is used.
One line is displayed for each unique memory range of the process.
The first field of each line shows the starting address of the memory range.
This field allows a correlation with the contents of the
/proc/<pid>/maps
file,
which contains the end address of the range and other information,
such as the access permissions and sharing.
The second field shows the memory policy currently in effect for the
memory range.
Note that the effective policy is not necessarily the policy
installed by the process for that memory range.
Specifically, if the process installed a "default" policy for that range,
the effective policy for that range will be the process policy,
which may or may not be "default".
The rest of the line contains information about the pages allocated in
the memory range, as follows:
N<node>=<nr_pages>
The number of pages allocated on
<node>.
<nr_pages>
includes only pages currently mapped by the process.
Page migration and memory reclaim may have temporarily unmapped pages
associated with this memory range.
These pages may only show up again after the process has
attempted to reference them.
If the memory range represents a shared memory area or file mapping,
other processes may currently have additional pages mapped in a
corresponding memory range.
file=<filename>
The file backing the memory range.
If the file is mapped as private, write accesses may have generated
COW (Copy-On-Write) pages in this memory range.
These pages are displayed as anonymous pages.
heap
Memory range is used for the heap.
stack
Memory range is used for the stack.
huge
Huge memory range.
The page counts shown are huge pages and not regular sized pages.
anon=<pages>
The number of anonymous page in the range.
dirty=<pages>
Number of dirty pages
mapped=<pages>
Total number of mapped pages, if different from
dirty
and
anon
pages.
mapmax=<count>
Maximum mapcount (number of processes mapping a single page) encountered
during the scan.
This may be used as an indicator of the degree of sharing occurring in a
given memory range.
swapcache=<count>
Number of pages that have an associated entry on a swap device.
active=<pages>
The number of pages on the active list.
This field is only shown if different from the number of pages in this range.
This means that some inactive pages exist in the memory range that may be
removed from memory by the swapper soon.
writeback=<pages>
Number of pages that are currently being written out to disk.
NOTES
The Linux NUMA system calls and /proc interface are only available
if the kernel was configured and built with the
CONFIG_NUMA
option.
Library Support
Link with
-lnuma
to get the system call definitions.
libnuma
and the required
<numaif.h>
header are available in the
numactl
package.
However, applications should not use these system calls directly.
Instead, the higher level interface provided by the
numa(3)
functions in the
numactl
package is recommended.
The
numactl
package is available at
ftp://oss.sgi.com/www/projects/libnuma/download/.
The package is also included in some Linux distributions.
Some distributions include the development library and header
in the separate
numactl-devel
package.
This page is part of release 3.14 of the Linux
man-pages
project.
A description of the project,
and information about reporting bugs,
can be found at
http://www.kernel.org/doc/man-pages/.
