[slurm-users] Only one socket for SLURM

[Chris Samuel]

On Thursday, 21 February 2019 1:00:52 PM PST Sam Hawarden wrote:

> Linux assigns numbers to your CPUs. 0-15 will be socket 1, thread 1. 16-31
> are socket 2, thread 1, 32-47 are socket 1, thread 2. 48-63 are socket 2,
> thread 2.

This isn't strictly true, for x86 for instance the kernel will read firmware 
tables to discover how the vendor has assigned the core numbering.

This means it can vary from vendor to vendor and model to model plus it can 
potentially change on a firmware upgrade, just to make life even more 
exciting.

This is why hwloc (and its predecessors) exist, to give a sane interface and 
default logical layout.  Slurm uses a similar system that results in something 
that looks very similar, so to Slurm CPU 0 is socket 1, core 1, thread 1 and 
CPU 2 is socket 1, core 1, thread 2, etc...

All the best,
Chris
-- 
  Chris Samuel  :  http://www.csamuel.org/  :  Berkeley, CA, USA