[slurm-users] 4 sockets but "

[Diego Zuccato]

Il 20/07/2021 20:30, Ole Holm Nielsen ha scritto:

Hi Ole.

> OK.  Performance may be a bit higher with SNC enabled.
I'll try to tune performance once it starts working :) BTW tks for the hint.

> Uh, that's an old Slurm which will have many bugs that are fixed in 
> later releases.  It seems that a number of sites use the very old Debian 
> distribution packages rather than modern Slurm versions :-(
That's a manpower issue: I'm alone managing all our servers, including 
the clusters. I'm noSlurm expert, nor MPI expert, nor storage expert, 
nor Infiniband expert, nor DB expert, nor many other things that would 
be needed to recompile Slurm from sources including the surrounding 
tools and libraries. The only possibility is to delegate that burden to 
the distro maintainers. Perfect? No. But it does the job. And usually 
the researchers are not interested in the latest features: if keeping a 
4-yo SO means they have a bit more time for their simulations, they 
press to NOT update...

> In 20.02 there was a bug with Boards and SocketsPerBoard, don't remember 
> if that was a problem in 18.08 as well.  See the link below for references.
The linked bugs seems related to a regression in later versions...

> Maybe delete Boards=1 SocketsPerBoard=4 and try Sockets=4 in stead?
Already tried. Actually, it's been the first try.

> The pam_slurm_adopt is very useful :-)
IIUC only if you allow users to connect to the worker nodes. I don't. :)

-- 
Diego Zuccato
DIFA - Dip. di Fisica e Astronomia
Servizi Informatici
Alma Mater Studiorum - Università di Bologna
V.le Berti-Pichat 6/2 - 40127 Bologna - Italy
tel.: +39 051 20 95786