Hi all,

I want to ask you about numad with a multi-processor program.

I have two machines.
First one has 16 cores and divided into 2 sockets(8 cores in 1 NUMA, 2 NUMA in CPU).
Second has 24 cores and divided into 4 sockets(6 cores in 1 NUMA, 4 NUMA in CPU).
The one NUMA has 400GB local memory.
The test takes about 300GB memory and runtime is about 15 hours with 8 processors.

Before using numad, the runtime variation was too large to allow the experiment to be proceeded.
After using numad, everything went very well and I used it very gratefully.

But the problem occurs when changing machine from 16 cores(8 in NUMA) to 24 cores(6 in NUMA).
As I mentioned earlier, the test works on 8 processors, but the process is not getting out of one NUMA with 6cores(rarely takes 2 NUMA).
The target utilization(-u option) is 100% and local memory is sufficient because numad treats inactive file cache as available memory.
Tested numad versions are 20150602(RHEL 6.9) and 20130804(RHEL 6.5).

Do you have any idea about this?
I'll very appreciate if you give me any solution.