'Introduction to HPC' workshop for Life Science in Innsbruck
Alper Yegenoglu and Michiel van der Vlag of the JSC's Simulation and Data Lab (SDL) Neuroscience held an "Introduction to HPC" workshop at the Life Science PhD meeting from 12-14 April 2023 in Innsbruck https://biomed-phd.i-med.ac.at/life-science-phd-meeting/, which was supported by the Human Brain Project's Education program. The workshop introduced the state-of-the-art of high performance computing (HPC) to scientists who had little or no background in working with supercomputers or compute clusters. The aim of the workshop was to teach what an HPC system is and how it differs from a regular computer, how an HPC system can help solve problems, how to submit and manage jobs on an HPC system using a scheduler, transfer files and use software through modules or in a virtual environment. Participants should also get a better understanding of the different types of nodes, storage partitions and other compute architectures.
The 25 participants of the workshop had backgrounds in a broad range of Life Science disciplines such as immunology, cardiology, lipidomics, metabolism, cell biology and neuroscience. Many of them were only somewhat familiar with HPC and had some basic level of programming experience, mostly in R or Python. The workshop started with a general introduction to HPC, its history, the concepts of computing time projects and quotasas well as some basic information on how to access the JUSUF HPC system at the JSC. After onboarding everybody via the JSC’s JuDoor user management system, each participant had access to a training account and could login to the Jupyter cloud on JUSUF. The handson session started by executing some example steps in a terminal from within this cloud. The decision to use the Jupyter Hub on JUSUF was made to be independent of the users’ individual operating systems. Additionally, notebooks had been prepared with exemplary code to execute.
The participants had to execute two different use cases related to research done at the SDL Neuroscience. First, the students worked on the Mountain Car problem, a computational task to push a car, positioned on a valley, up a mountain to a certain height. The openAI gym/gymnasium environment, offering an interface to this problem, was connected to a biologically inspired, spiking neural net implemented with the spiking neuronal network simulator NEST. The network’s connection weights were optimized with a genetic algorithm using the Learning-to-learn (L2L) framework, so that the network was able to push the car. The students learned how to use and play around with a machine learning library on a single HPC node.
The second use case entailed a grid search parameter sweep for a computationally demanding simulation, using The Virtual Brain (TVB). In this simulation a reference signal is fitted, utilizing a GPU and similarly a multi-node CPU implementation using MPI, to an earlier obtained example data set. In the latter use case the students could tune the parameters of the GPU TVB simulation and enlarge the number of nodes involved.
The workshop went very well as all participating students were able to at least execute the predefined exercises, much of which could also be ascribed to the experience gained from a previous similar workshop held at the HBP Summit 2023 in Marseille.