Summer of HPC comes to an end
Posted: 2 Sep 2016 | 11:08
These students from around Europe have spent the last 7 weeks with us at EPCC immersed in HPC, and each working on a specific project in the field. This is a great because not only do they gain experience and interest in HPC but we also get a useful, tangible, outcome from these projects.
Visualising the weather
I have been mentoring Tomislav and together we have been developing a weather demo for outreach use, where the public can set up certain weather conditions and then see how the weather would proceed from there.
It is based on the MONC cloud model that I have been developing with the Met Office and this model then runs on Wee Archie, our mini supercomputer made out of Raspberry Pis. These initial conditions are submitted to Wee Archie, which runs the simulation and as the state (such as cloud, rain, temperature etc) of the model progresses it is dumped out in real time which is then displayed on screen. From a visualisation perspective it looks like a movie of the weather and one can really see changes in the weather as it is being simulated.
Whilst this was initially targeted at the general public, we realised part way through that it was also a great opportunity to target more knowledgeable audiences and could even act as an educational tool for undergraduates. Based upon Tomislav’s work it is very easy to tweak more in-depth model parameters and then get immediate feedback as the model is running on exactly what impact these changes have on the system.
For instance there are two different advection schemes which compute movement through the atmosphere due to wind. From a computational view we can describe one as energy conserving but requiring considerable computation time, and another as non-conserving but much faster. However being able to select either one, view the results in real time and then select the other for comparison provides a very obvious illustration of the difference that these make to the actual simulation rather than just comparing the raw numbers.
From an HPC perspective the user can select the number of processes in both dimensions and the number of cores per node to utilise. Not only does the application illustrate how the area is split up amongst the different Wee Archie cores, but also a plot of communication vs computation for each core is displayed in real time. This allows us to see the impacts of strong scaling, where on smaller core counts (for the problem size) computations dominate but as we increase the number of cores communications start to become more significant. It is also possible to see the impact of uneven decomposition, which is when some cores have more work than others to complete.
This is the first visualisation of MONC data and Tomislav has done a great job on the project. His work acts as a foundation for future developments and we are also planning to use this at an outreach event (Bang Goes The Borders) in late September.
Summer of HPC
This is the fourth year I have been a mentor on the SoHPC programme and not only do I think it is a fantastic way of getting people involved with the field and experience in developing codes, but I am also always constantly surprised by the quality of work that participants produce and what they manage to achieve in their 7 week projects.
Watch the video
Nick Brown, EPCC