Drought - Getting Started Guide

Drought is a small 4-node temporary test cluster set up to test NWP24 hardware and software integration. Note that it has some very specific quirks that will cause you and your collegues pain if you don’t read through this guide.

Hardware

The nodes have 2 x AMD EPYC 9565 72-Core Processors and 768GiB RAM.

Threading is turned on, which gives you 288 virtual cores per node.

Do not use the login node for almost anything besides copying files

The login node “drought1” should not be used when compiling. At is of a much older hardware type than the NWP24 hardware and will produce potentially broken binaries.

The login node “drought1” should mostly be avoided when running programs. Anything you compile for this cluster or run from the /software NSC compiles might be broken.

Instead log in to the first compute node with “ssh -p 222 n1” and do all your compiling and executing there.

You can also ssh to the three other nodes in the same way due to a technicality, but that would disturb actual jobs. Do not do it!

Using slurm

All users run under the same Slurm project: nwp24-pilots

This is the default account that is choosen automatically by sbatch and interactive, but if you have script with hardcoded accounts like “-A metcoop” they need to be updated.

Use the few nodes available responsibly

There are at the best of times three nodes avilable to run jobs since one is dedicated to compiling. Do not start long jobs taking up all nodes before coordinating that within the group.

Note that the new nodes are much faster and can be shared. Try using -n to allocate a number of cores for your test instead of allocating the whole node.

Available filesystems

/nobackup/prod{1,2}: The same filesystems that are mounted on Stratus, but they are read-only mounted to prevent experiments from disturbing production.

/home: This is a separate and temporary /home with a much higher user quota than usual. This is to let you both install binaries in your homes and copy datasets there if they are not available via /nobackup from Stratus.

/software: A freshly compiled installation of the software NSC provides. This is very bare bones right now, but contains the basics to get you compiling your own environments and tests. Let smhi-support@nsc.liu.se know if you are missing something.

Getting a login account

See the SMHI Access page more information.

How to log in

See the general instructions on getting started for information on how to login for the first time and set your permanent password.

• The Drought login node is drought1.nsc.liu.se.

The SSH key fingerprint for the login node is:

SHA256:qotif1uTYDgfUYARirhgU6fLSLTTyVEH55DsHSrOUog drought login node (RSA)