numaExplanation.txt

Notice we are using specific options with numactl to start up kdb 
The general recommendation when using numa is to set --interleave=all - from https://code.kx.com/q/kb/linux-production/ 
` numactl --interleave=all q `

However this isnt always the most optimal option for all cases. Lets use simple insert to test the time to allocate memory.
In laymans terms numa is arranged into nodes. And cores can access memory in the same node faster. Taking this simple 2 node example. 
We can see that if we use cores from a different node to memory we can see a minor performance hit.

```
|22:26:51|virtu@clx4:[~]> numactl --interleave=all q -q 
q)\ts:1000000 a,:1
111 12583152
\\
|22:27:05|virtu@clx4:[~]> numactl -N 0 -m 0  q -q 
q)\ts:1000000 a,:1
111 12583152
\\
|22:27:43|virtu@clx4:[~]> numactl -N 0 -m 1  q -q 
q)\ts:1000000 a,:1   
115 12583152
\\
|22:28:02|virtu@clx4:[~]> numactl -N 1 -m 0  q -q 
q)\ts:1000000 a,:1
114 12583152
\\
```

Whenever we create the persistent memory mounts they too like dram itself have an associated node. And we will achieve the best performance by running the q process in the correct node for the associated mount provided to -m  
Example of memory hit from

```
|22:28:11|virtu@clx4:[~]> numactl -N 0 -m 0  q -q -m /mnt/pmem0/
q)\ts:1000000 .m.a,:1
188 704
\\
|22:28:54|virtu@clx4:[~]> numactl -N 0 -m 0  q -q -m /mnt/pmem1/
q)\ts:1000000 .m.a,:1
200 704
\\ 
|22:29:22|virtu@clx4:[~]> numactl --interleave=all q -q -m /mnt/pmem1/
q)\ts:1000000 .m.a,:1
200 704
\\
```