HPC@Sheffield, the University of Sheffield's annual conference for researchers using HPC, took place on Monday 8th April. The programme saw researchers, industry professionals, and research IT specialists discuss the ever-changing face of computationally intensive research, challenges faced, and new ideas to meet these. Quite apparent was the breadth of disciplines represented, with speakers from fields as diverse as experimental psychology and in silico medicine research highlighting that HPC is no longer the reserve of the physical sciences and engineering.
The need for unique architectures to address certain problems was demonstrated in the keynote address regarding the SpiNNaker platform. Part of the Human Brain Project, this involves an entirely novel architecture intended to model a subset of the human brain in real-time using a fraction of the energy a typical supercomputer would require to perform the same task. Following the breakdown of Moore's law and Dennard scaling, lessons learned from such projects may well prove vital to meeting the constantly increasing demand for computational resources.
Another challenge faced is dealing with the vast amounts of data continuously collected and produced from myriad sources. High Performance Data Analytics has emerged at the interface of HPC and data intensive computing. An invited talk from Dell HPC attempted to define the obstinately indefinable “big data” (a dataset large enough to crash your workstation is an apt description!) and to elucidate some of the techniques for processing it, but also warned of the dangers of bias in data and models.
While Cloud computing has been embraced across the enterprise IT, HPC has in general been slow to adopt this approach. An invited talk from Microsoft Azure focused on the HPC-centric resources that are now available to researchers, demonstrating that vendors are eager to meet the unique needs of the sector. Researchers from the Insigneo institute also discussed the use of Cloud APIs in facilitating the sharing of data.
A common theme highlighted by many of the presenters was the requirement for reproducibility of computational research. Some high-profile retractions in recent years due to erroneous results have inspired new methodologies, procedures and indeed disciplines. Research Software Engineering has emerged as a distinct and valuable role in applying professional code development practices to academic software. RSEs from the University of Sheffield spoke about the benefits to researchers of using these services and introduced The Turing Way, which is being developed as a guide to improving reproducibility.
The day was rounded of with presentations for the University research software competition. This placed considerable focus on the methodologies employed in the development of each project, with a view to emphasising the importance of robustness and reproducibility.