Interactive HPC Computation with Open OnDemand and FastX

As dataset size and complexity requirements grow increasingly researchers need to find additional computational power for processing. A preferred choice is high performance computing (HPC) which, due to its physical architecture, operating system, and optimised application installations, is best suited for such processing. However HPC systems have historically been less effective at the visual display, and least of all in an interactive manner, leading into a general truism of "compute on the HPC, visualise locally".

This is primarily due to the tyranny of distance, but also with additional latency introduced by contemporary graphics when remote display instructions are sent to a local X-server. With a demand for both HPC computational power and interactive graphics, the University of Melbourne has implemented two technologies, FastX and Open OnDemand, on their general-purpose HPC system "Spartan". This allows users to run graphical applications on Spartan, by submitting a job to the batch system which executes an XFCE graphical environment. In the Spartan environment, FastX has been coupled with Open OnDemand which provides web-enabled applications (e.g., RStudio, Jupyter Notebooks). In illustrating how this environment operates at the Spartan HPC system, the presentation will also illustrate recent research case studies from the University of Melbourne that have utilised this technology.

A presentation to eResearchNZ 2021