I am a PhD student in the Department of Mathematical Sciences, coming to the end of my first year of research. I work with Dr Eike Müller and Prof Ivan Graham. My PhD title is:
"Preconditioners for Higher-order Discontinuous Galerkin discretisations of elliptic PDEs on modern architectures and applications in atmospheric modelling"
My interest in HPC started during my undergraduate degree, using the Minerva supercomputer at the University of Warwick. It is only fitting that I continue my research interest at the University of Bath on their supercomputer Balena, which enables me to run my simulations over even more computers than before.
In June 2017, I was lucky enough to participate in Bath's sixth annual HPC symposium. The symposium gave me a brilliant chance to showcase some of my research and demonstrate how I was using the HPC facilities here at Bath. My focus at the symposium was on the "modern architectures" part of my title. Over the course of the PhD we want to look at the feasibility of using modern mathematical techniques for numerically solving partial differential equations (PDEs), specifically the equations that are used to predict the weather. Even today, huge computers are required to forecast and model the atmosphere, this research investigates whether we can do so:
- More accurately
- In the same time frame (or faster!)
- Without using more energy than is needed
- Whilst making full use of the architecture available
- Using Higher-order Discontinuous Galerkin methods
My presentation included a brief introduction to precisely what this entails.
Throughout the day we saw many excellent presentations from across the university, the day was coming to a close, my presentation approaching, and all of the preceding presentations had been of a high standard. I really had to pull something out of the hat to get the audience interested in my research. Maybe I could use the Balena visualisation service to knock their socks off! So, I demonstrated the visualisation (using Balena) of a pre-computed solution to a PDE on the surface of the Earth. You can see my slides and some of my research on my website, http://people.bath.ac.uk/jdb55. I look forward to trying out these techniques on Balena's new hardware as part of its mid-life upgrade.
I would encourage anyone across the University of Bath who finds themselves running long simulations (especially on a laptop), to invest some time learning to use Balena to offload this work, as it can significantly improve your work flow. My thanks go to the organisers of the symposium, ClusterVision, and the tireless work of Steven and Roshan.