[cfarm-users] Correct number of CPU/cores/threads (Was: Setting up GitLab CI for git.git on the farm)

[Segher Boessenkool]

Hi!

On Thu, Dec 20, 2018 at 12:49:59PM +0100, Baptiste Jonglez via cfarm-users wrote:
> So, I just added pretty usage bars [1] to the list of machine :)

Thanks.  But maybe it should be presented differently?

The page now is much too wide to fit on a screen, and those little bars
are all jumbled around, so it is hard to get information from this.  The
links ("CPU graphs etc.) distract a lot from the information in that
column, too.

> Since I needed the correct number of cores/threads to compute a percentage
> of CPU usage, I took the opportunity to manually fix the cases where
> Ansible gets it wrong:
> 
> - EdgeRouters: gcc22, gcc23, gcc24
> - Mustang aarch64 boards: gcc113, gcc114, gcc115, gcc116
> - Opteron aarch64 boards: gcc117, gcc118
> - POWER machines running Linux: gcc110, gcc112, gcc135
> 
> If you see any other wrong CPU information in https://cfarm.tetaneutral.net/machines/list/
> I can also fix it.  For instance, the number of cores/threads for the AIX
> machines looks incorrect, but I have no idea how to get the correct
> information (does hwloc even build on AIX?).

I normally use
  lscfg -vp | grep proc
(no idea how good *that* is, but it does the job).

gcc111 is 12 cores, 48 threads; gcc119 is 16 cores, 128 threads.

> I realize now that "CPU" vs "cores" is a simplistic distinction for some
> machines.  I tend to take the definition "CPU = physical socket" but that
> sounds incorrect for the aarch64 machines: do they really have 8 CPU
> sockets as reported by lscpu?  Or is it just that each core has its own L3
> cache so it can be considered to be an independent CPU?

Many people use cpu = core.  Linux uses cpu = thread.  Some people use
cpu = die; and some (like you) use cpu = package (not everyone has sockets).

L3 cache topology does not have much to do with cpu topology.

Btw, gcc110 has conflicts on the munin-node package since the update
to Centos 7.6, so no pretty graphs there :-/


Segher