This suite contains a set of simple programs to test numa interfaces provided by libnuma.
Make sure you have numactl, if you don't, do
sudo yum install numactl
sudo yum install numactl-devel
or use other package package management tool for your platform.
To check if your machine has NUMA, do
numactl --hardware
The first line shows the number of NUMA nodes in the system. It needs to be more than one node.
Alternatively,
sudo ls /sys/devices/system/node/
Directory node0 represents node 0, 'node1' for node 1, etc. If you only have node0, then there is only one node.
Alternatively, write a simple program to call numa_available(), if it returns -1, all other functions in this library are undefined.
The Linux kernel will use the INTERLEAVE policy by default on boot-up to avoid putting excessive load on a single memory node when processes require access to the operating-system structures. The system default policy is changed to NODE LOCAL when the first userspace process (init daemon) is started.
Node is frequently used in the document, more formally, a node is defined as an area where all memory has the same speed as seen from a particular CPU. A node can contain multiple CPUs. Caches are ignored for this definition.
A good overview
cd make/; make all
To make a single test case, do make move-pages, where move-pages is the test case that tests move_pages(). Cat the makefile to see all the targets.
To run the test, do cd ../build/; ./move-pages --help.
You may use pcm-memory.x from Intel PCM to verify if you are really using a NUMA node. Caveat: create enough memory bandwidth to make monitoring obvious.