The onload_stackdump lots command is used to verify that allocations occur on the required NUMA node:
# onload_stackdump lots | grep numa
numa nodes: creation=0 load=0
numa node masks: packet alloc=1 sock alloc=1 interrupt=1
The load parameter identifies the node where the adapter driver has been loaded. The creation parameter identifies the node allocating memory for the Onload stack. The numa node masks
identify which NUMA nodes allocate memory for packets and for sockets, and the nodes on which interrupts have actually occurred. A mask value of 1 identifies node 0, a value of 2 identifies node 1, a value of 3 identifies both nodes 0 and 1 etc.
For most purposes it is best when load and creation identify the same node which is also the node local to the network adapter. To identify the local node use the following:
# cat /sys/class/net/<interface>/device/numa_node
The cpu affinity of individual Onloaded threads can be identified with the following command:
# onload_stackdump threads