We're taking advantage of our parent university (CMU)'s site license for Red Hat Linux to migrate to RHEL 8.x. - Derek --- Derek Simmel Pittsburgh Supercomputing Center +1 (412) 268-1035

It looks like the ssh keys were updated/changed on your host, perhaps as a result of your (re)installation. The initial warning comes up because your ssh client has StrictHostKeyChecking enabled, and

#openhpc

Hi Yun, Which OpenHPC recipe (installation instructions) are you using? Are your nodes configured for EFI or Legacy boot? Warewulf conf files are normally found under /etc/warewulf; the dhcpd.conf fil

Yun, Ideally, you should have your BMC connections separated from your internal network, since access to the BMC network by users presents a security risk. To do this on a single switch, the switch wi

Yun, Using 'ipmitool lan set access on' on the sms node does that on the sms node's BMC only. You need to be able to reach the BMCs on the compute nodes from the sms node to control those nodes. Do yo

Yun, Can you ping the bmc addresses successfully from your sms node? I don't see a password being supplied (usually -P <password>) to your ipmitool command - did you set one on your compute node BMCs?

I agree with Alan on this - and adding Ubuntu is an entirely separate effort. If you're going to switch horses, the alternative we have available is OpenSLES Leap 15. Meanwhile we need to gather facts

Patrick, Jacob's prior question was about utilizing idle cores on nodes. With the slurm.conf parameters as you listed them (and OverSubscribe=NO, which is the default), CPUs (rather than cores per CPU

Jacob, I always have to try out each configuration change one at a time until I see the behavior that I need between the scheduler and the jobs that I'm running. If you can tell us a bit more about wh

Hi Jacob, There are a variety of ways to approach this in your SLURM configuration. If you set up your Partition(s) with OverSubscribe enabled, then jobs can share the resources (nodes) allocated. See

Jacob, Our local guru suggests looking at SLURM's heterogeneous job docs at https://slurm.schedmd.com/heterogeneous_jobs.html Essentially you would tell the job what resources to ask for and how many

Jacob, I doubt your bios' have anything to do with it. This is more of a SLURM setup and sbatch options equation to solve. I'll consult our SLURM guru here for clues. Others please chip in if you have

Jacob, Your srun command asks SLURM to allocate 3 nodes and to run 20 tasks on them, comprised of running /bin/hostname. Note that these tasks are independent of each other, and may not run in paralle

OK, so you can get to them all - that's good. And what, if anything is listed when you run 'squeue' ? - Derek --- Derek Simmel Pittsburgh Supercomputing Center dsimmel@... +1 (412) 268-1035

Given p[0-2], and assuming thy are all available, what do you get if you try a simple 'srun -N 3 /bin/hostname' ? Do all the nodes' hostnames show up? - Derek --- Derek Simmel Pittsburgh Supercomputin

Jacob, What does your PartitionName=putnam configuration line in your slurm.conf look like? - Derek --- Derek Simmel Pittsburgh Supercomputing Center dsimmel@... +1 (412) 268-1035

The RealMemory values look wrong to me - what do you get on the nodes when you run 'slurmd -C' to show what slurmd thinks they have? - Derek --- Derek Simmel Pittsburgh Supercomputing Center dsimmel@p

Jacob, In your earlier e-mail, the command you wrote had --partition=putname instead of --partition=putnam - Derek --- Derek Simmel Pittsburgh Supercomputing Center dsimmel@... +1 (412) 268-1035

Does fdisk work? --- Derek Simmel Pittsburgh Supercomputing Center dsimmel@... +1 (412) 268-1035

We make it available for our production HPC systems at PSC. There's an investment you have to make in setting it up; instructors with students seem to find it most useful. Note that users should use C