For me, it doesn't: [root@cn45 ~]# slurmd -C ClusterName=(null) NodeName=cn45 slurmd: Considering each NUMA node as a socket CPUs=64 Boards=1 SocketsPerBoard=4 CoresPerSocket=8 ThreadsPerCore=2 RealMemory=128832 TmpDisk=32768 UpTime=7-01:47:00
[root@cn45 ~]# grep 'NodeName="cn\[45-49\]"' /etc/slurm/slurm.conf NodeName="cn[45-49]" NodeHostname="cn[45-49]" Sockets="2" CoresPerSocket="16" ThreadsPerCore="2" RealMemory="128832" Weight="1000" Feature="intel,haswell,supemicro,nvme" [root@cn45 ~]# In any case, I found this thread that seems to answer my question: https://bugs.schedmd.com/show_bug.cgi?id=838 Ryan On Thu, May 19, 2016 at 2:36 PM Ryan Novosielski <[email protected]> wrote: > Bear in mind that that stuff is set in > > On May 19, 2016, at 4:29 PM, Ryan Braithwaite <[email protected]> wrote: > > > > Hi, > > > > I just upgraded our cluster to 15.08.11 and started using node features > to constrain jobs. We discovered that our Intel Haswell-based systems that > have Cluster-on-Die enabled are showing up with the wrong number of sockets: > > … > > > > I'm not sure if this is a known bug or not, so I thought I'd check here > first. > > Bear in mind that unless I’m misunderstanding you, that stuff from > scontrol show node is set in slurm.conf. If you want to check what SLURM > thinks you ought to have in slurm.conf, you can do “slurmd -C” from a node. > > -- > ____ > || \\UTGERS, |---------------------------*O*--------------------------- > ||_// the State | Ryan Novosielski - [email protected] > || \\ University | Sr. Technologist - 973/972.0922 (2x0922) ~*~ RBHS Campus > || \\ of NJ | Office of Advanced Research Computing - MSB C630, Newark > `' > >
