What machines we have and what they are for

What machines we have

You can find more about the diferent machines that make up out cluster by running the

s-hosts

command. It lists all the machines according to their partition and shows relevant information regarding the machines.

Partitions

CPUs

The amount of CPUs available for the slurm to manage. This doesn’t necessarily correspond to the total amount of CPUs of the machine, specially for those located in the workspace (otherwise there would be no more cores to dedicate to all the other tasks you are doing right now, including having this window open).

Load

Shows the real time average load of the machine. This is the amount of actual cores being used. For instance a value of 6.5 should be interpreted as 6.5 cores in use. This could be due to the machine running a 6 core slurm job along side other processes (running outside of slurm) that only take half a core. For machines in the workspace it is possible for the load to exceed the amount of available CPU cores shown, but remember that those are only the dedicated slurm cores, it is likely the machine posses more than the cores shown.

Mem(GB)

Shows the RAM allocation (in GB) for each machine. The RAM requirements of a job depends, regardles if a machine has no memory available it is not recomended to run additional jobs.

Cluster status

If this information is being show the machine is properly running as part of slurm, but that is not always the case. Some machines may have some hardware issue and be down for repair, have been rebooted or be offline all together. These statuses obscure the system resources because they are not in the slurm queue and thus are not being monitored. In order to get the real occupation you need to tunnel into the machine (ssh bioXXX) and analyse the top output. You will find that it provides more information than s-hosts, including the information on load and CPU usage discussed previously.

What are the machines for

GPU machines should not be used to run analysis

The machines with the names bio001 up to bio150 are fitted with GPUs and should be exclusively used to run GPU jobs through the slurm. As most of these nodes also double as workstations, the resources of these machines are already stretched thin and should not be overloaded with additional jobs (be them initializations, minimizations or analysis). ## Nodes with names over bio151 should be used to run init, min and analysis instead You will eventually find you need to run jobs outside of the slurm, be them initialization, minimization or other analysis. For these you can use any machine with a name over bio151. Just make sure to consult

• s-hosts to ensure the machine has available cores (the fewerr occupied cores the better); and
• rtop bio### to make sure there are no other non-slurm jobs running. This is because you should make sure to not run more than one init/min/analysis per machine. For analysis that could take advantage of GPU acceleration you should take advantage of GPU nodes outside the queue (flagged for DRAIN for example) and avoid running them in your local workstation