Jupyterhub on an on-premises cloud – a special focus on gpu accelerated machine learning and 3d visualization

Abstract

At the IT department of the University of Münster (WWU IT) we build a private IaaS cloud based on OpenStack and Kubernetes (WWU Cloud). This cloud provides a generic platform for data storage and service hosting. WWU IT operates a JupyterHub on WWU Cloud for use in research and education. Researchers have access to virtual GPUs from their Jupyter sessions. These may be used to compile and natively run CUDA accelerated code, e.g. for machine learning. Using VirtualGL, we also provide an accelerated X server in Jupyter sessions. X11 applications are then accessible from the browser using noVNC.