HPC manual
Getting access
To receive access credentials, you must fill in the application form. After receiving the application, the HPC Centre will review it and contact you to inform you on the following steps. You will have to sign a User Agreement.
General guidelines for using the HPC system
Information regarding general principles that should be followed when working with the cluster, as well as information about technical support.
HPC hardware specifications
Information regarding system parameters, access network, computing cluster Rudens, data storage, as well as nod names and parameters.
Software installed on the cluster
A summary of software installed on the HPC cluster that will be available wafter you become a user of the cluster.
User interfaces
Information about the command line, graphical interface of the software, and web interface.
Getting started with the cluster
Information about connecting to the cluster, user workspace, copying files, as well as using software modules.
Job management
This section contains the following information:
- Queuing a job;
- Job queues;
- Monitoring queues, nodes, and jobs;
- Job efficiency;
- Best practices for the jobs with intensive input/output (I/O);
- Transition between Commands Torque/PBS Pro and Slurm.
Using GPU’s and CUDA on the cluster
Information on requesting GPU’s for a job and development using CUDA Toolkit – preparing the working environment, usage examples for the cluster, and GPU code generation options.
Running containers in the cluster
Information on singularity that is a containerization solution designed for HPC cluster environments. Native singularity containers can run also docker containers.
Usage accounting
Information on obtaining a balance sheet and detailed report on the use of HPC services.
Useful online resources
- RTU HPC Center’s website: http://hpc.rtu.lv
- Job management system (Torque/Moab) documentation: http://www.adaptivecomputing.com/support/documentation-index/
- An online magazine about HPC technologies: https://www.admin-magazine.com/HPC
Parallel commands with pdsh
PDSH executes a parallel shell command on multiple nodes (from a login node), for example,
- Check the load of the cluster nodes (load average)
pdsh -w wn[01-11,43,48-60] uptime- Check the contents of scratch directory on each of the cluster nodes
pdsh -w wn[01-11,43,48-60] ls -la /scratch | grep Conda/Anaconda environment
The package and environment management system allows users to create isolated working environments, install various programming tools, including the recent Python versions, ML frameworks, and bioinformatics tools.
Preparing the working environment
module load condaUsage examples for the cluster
module load conda
conda create -n my_conda_enev python=3.9
conda info --envs
source activate my_conda_env
conda install pytorch torchvision -c pytorch