HPC workshop

Workshop: Next Generation HPC architectures for studying climate variability

20th - 22nd March 2017, Exeter.

** Summary output paper:

Next-Generation High Performance Computing for the Environmental Sciences: a Roadmap

This ReCoVER meeting was co-organised by Beth Wingate (University of Exeter), Lawrence Mitchell (Imperial College, London) and Chris Budd, (University of Bath).

The purpose of this meeting was to imagine and build a roadmap for taking advantage of next-generation HPC computer architectures for studying climate variability. This is thought to be a serious issue since most parallelisation strategies used in state-of-the-art climate models have the following issues: 1) the number of cores is nearly optimal for today’s grid resolutions – adding more compute cores no longer decreases the time-to-solution and 2) increases in grid resolution require longer simulation times due to the CFL criteria. Therefore, current climate models will have to be retooled to improve both numerical methods and compute/software infrastructure to make use of more parallelism.

At the present moment, most climate models will scale well on machines thought to be suitable contenders for ARCHER 2, but beyond that less is known both about the possible architectures and strategies for ‘going beyond the strong scaling limit’. Some areas of discussion included:

1. What sort of novel numerical algorithms might be developed (and applied to environmental modelling) that fully exploit emerging architectures, and what are the challenges to numerical analysts from these algorithms?
2. While the longer-term processes of item 1 are ongoing throughout the next decade, what can we do to make best use of these architectures? For example, should we concentrate on more ensemble runs, more statistics/machine learning, or diverse model runs at the same time? Are there new science questions that can be asked that can have an impact on UK and international science?

We had a small number of keynote speakers about the future of computer architectures, what is currently planned, followed by a workshop-style meeting where participants were encouraged to engage across disciplinary boundaries to devise a strategic path forward for research that will make best use of the new architectures.

Schedule:

Monday 20th March 2017, Innovation Centre lecture theatre, University of Exeter

• 12:00 lunch
• 13:20 Introduction
• 13:30 Lecture: Lawrence Mitchell (Imperial)
• 14:15 Lecture: Katherine Evans (US Department of Energy)
• 15:00 Tea
• 15:30 Lecture: Matthew Chantry (Oxford)
• 15:50 Lecture: Chris Maynard (Met Office)
• 16:10 Lecture: Willem Deconinck (ECMWF)
• 16:30 Lecture: Hans Vandierendonck (Queen's University Belfast)
• 17:00 Close
• Evening: self-organized meal in Exeter

Tuesday 21st March 2017, Innovation Centre lecture theatre, University of Exeter

• 9:15 HPC challenges: Beth Wingate (Exeter)
• 9:30 Lightning talks (7 minutes each): Richard Everson (TBC), Jemma Shipton (Imperial), Eike Mueller (Bath), Paul Dellar (Oxford), Dejice Jacob (Glasgow), Serge Guillas (UCL), Peter Challenor (Exeter)
• 10:45 break
• 11:10 Agenda-setting activities
• 12:30 Lunch
• 13:30 Walk
• 15:30-17:30 Discussions
• 19:00 Workshop meal, Sam Jones'

Wednesday 22nd March 2017, Innovation Centre lecture theatre, University of Exeter

• 9:00 Discussions
• 10:30 Coffee
• 11:00 Reporting back
• 12:30 Lunch
• 13:30 Close

Follow-on funding: This is a workshop organised by the EPSRC network ReCoVER. The network offered participants the opportunity to apply for a feasibility funding project of up to £25k to follow up ideas from the workshop.