Usage

This description intends to offer starting points for eX3 and how to use it. We will continuously improve this documentation.

Prerequisites¶

In order to access eX3, a user will have to file an application and register first. For that purpose please follow the registration process as documented here.

Once access has been granted, use the credentials and received login instructions to access the system.

Logging in¶

User can login from within Simula or from extern using dnat, where dnat will drop you on srl-login1:

    ssh username@dnat.simula.no -p 60441

Within Simula network users can use one of the available login nodes:

• srl-login1.ex3.simula.no (priority)

• srl-login3.ex3.simula.no

So in case srl-login1 should not be available (for whatever reason), srl-login3 serves as fallback.

Filesystem¶

A user has access to the usual home directory on the current login node. In addition, a user has access to shared storage on /global/D1/. In addition project specific paths can be created upon request.

Since reading from shared filesystem can be a bottleneck for data heavy jobs, a local (faster) NVME disk can be used. For that purpose, when running a job, prepare data (copy it over from the global share) into /work/<username>/ on the node that requires the data.

Getting an overview¶

In order to gain an overview over available resource on eX3 you can navigate to https://naic-monitor.simula.no - as long as you can access the internal network. Depending on how you access eX3 you might have to set up port forwarding ssh ... -L 443:443 .

Running an interactive job¶

Start an interactive job, e.g., requesting one gpu and minimum 2 cpus

    srun -p dgx2q --gres=gpu:1 --mincpus=2 --pty /bin/bash

Once you have gained access to the system, check that the GPU is visible, here for an NVidia GPU

    $> nvidia-smi -L
    GPU 0: Tesla V100-SXM3-32GB (UUID: GPU-ad466f2f-575d-d949-35e0-9a7d912d974e)

    $> echo $CUDA_VISIBLE_DEVICES
    0