Based on the previous ENES HPC workshops in Lecce (2011 & 2018), Toulouse (2013 & 2016) and Hamburg (2014), this year’s event had originally been planned as an on-site event for about 100 participants.
The workshop series is being organised by the “ENES HPC-Task Force” with workshops taking place roughly every two years. DKRZ hosted this year's workshop, which was funded by the Horizon 2020 project ESiWACE2. Due to the situation with COVID-19, the event was held as a virtual conference for the first time with approximately 180 participants from 12 European countries, the United States, China, Japan and India.
The workshop brought together scientists from the fields of earth system modeling, informing them about the latest developments in high performance computing (HPC) and fostering a mutual exchange, also involving HPC experts.
Talks were structured in four sessions focusing on:
- Very high-resolution modeling
- Performance portability
- Machine learning for parameterization schemes
- Challenges in exascale data processing and visualization
In order to make individual workshop days less exhaustive with presentations being delivered online, the tight original timetable was rescheduled: the 35 talks were spread over four days, each with approximately three hours of video conference. A morning session for presenters from Asia and afternoon sessions with contributions from Europe and the US allowed all speakers to present their work during local daylight hours.
All talks were held as a videoconference, whereby the presentation slides as well as the speaker were usually broadcasted live to all participants via screen sharing. Questions for the speakers were collected in online documents during the talk and directed by the chairpersons to the speakers afterwards. This approach had already been adopted for the general assembly of the partner-project IS-ENES3 and proved to be effective for the virtual ENES HPC workshop as well.
DKRZ staff organised the event and actively contributed to the workshop program. Dr. Florian Ziemen of DKRZ and Dr. Daniel Klocke of the German Meteorological Service (DWD) jointly chaired the first session, which focused on the analysis of very high resolution and preferably globally coupled storm- und eddy-resolving climate simulations as well as solutions for their design and initialization strategies.
In her talk themed „Interpretable Machine Learning“, Dr. Maria Moreno de Castro from DKRZ’s data management department advised caution when using so-called “black box models” for machine learning (ML). With such models, users cannot easily identify the basis on which an ML algorithm has made a decision. This may lead to unrecognised systematic biases in the results. As an alternative and in addition to interpretable “glass box models”, she introduced several methods of “explainable artificial intelligence” (XAI) that render decisions taken by the ML algorithm traceable and transparent. At the end of her talk she built a bridge from explainable models to their application for parameterizations of cloud formation and precipitation processes in weather and climate model simulations.
Dr. Niklas Röber, co-lead of the ESiWACE2 work package for data post-processing, analytics and visualization, contributed a talk on current and future activities in the field of large data visualization.
A highlight of the first session was the presentation by Prof. Bjorn Stevens, director and head of the department “Atmosphere in the Earth System” at the Max Planck Institute for Meteorology (MPI-M), on next generation earth system models. He examined the lessons learned so far from a technical, scientific and sociological perspective and outlined future challenges that call for an intensification of joint and coordinated efforts.
Prof. Satoshi Matsuoka, director of the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan and professor at the Tokyo Institute of Technology, kicked off the session about performance portability on the second day. He gave a thorough introduction to Fugaku, the new exascale supercomputer for Japan, which is currently being installed under the lead of R-CCS and shall replace the K-Computer in 2021. Design, development, system configuration and performance of the new exascale supercomputer as well as intended research activities were addressed in his talk. In addition to supporting earth system research, Fugaku shall, i.a., also be used to simulate the spread of COVID-19.
Two other contributions to the session on performance portability were made by ESiWACE2 project partners from the UK: Rupert Ford of the STFC Hartree Centre spoke about the latest developments in PSyclone, a domain-specific compiler program that is also used in LFRic, the next generation weather and climate model of the MetOffice. PSyclone efficiently translates domain-specific model code into code that is optimised for complex computer architectures. In the subsequent talk Dr. Iva Kavcic of the MetOffice illustrated how PSyclone is being used in the LFRic model, how that influences the model performance and which activities are planned for the future, e.g. porting LFRic to GPUs.
Representing the HPC industry, Stan Posey of NVIDIA Corporation contributed a presentation on the latest GPU developments and their applications for climate and weather simulations to the workshop.
With the event shifted from physical to virtual, the number of participants increased and the workshop was accessible to a broader audience, especially to those scientists who do not have funding to attend on-site events. Moreover, the carbon footprint of the event was drastically reduced due to the virtual format that required neither travel nor accommodation for the participants.
While the presentations and the associated discussion rounds were successful, the event nevertheless lacked the side-discussions and the exchange that are typically an important part of on-site meetings. New ideas are needed to foster mutual exchange and networking among the participants to make future low-emission events, i.e. virtual conferences and workshops, even more useful and beneficial.
On May, 27, the workshop was suspended for the ESiWACE2 annual meeting, which also was held as a videoconference and brought together 53 attendees from ESiWACE2 consortium member institutions. In addition to progress reports from the seven work packages, the full day meeting comprised discussions on merging workflows within the project, e.g. for simulations, data storage and data post-processing, as well as on sustainability and HPC initiatives.
The program committee of the 6th ENES HPC workshop consisted of Joachim Biercamp, Dela Spickermann and Florian Ziemen (all DKRZ), who were in charge of organizing and hosting the event, and of Daniel Klocke (DWD), Sophie Valcke and Jean-Claude André (CERFACS), Mario Acosta and Kim Serradell (BSC), Reinhard Budich (MPI-M), Peter Dueben (ECMWF), Sandro Fiore (CMCC) and Niklas Röber (DKRZ).
The 7th ENES HPC workshop is scheduled for 2022 and will be organised by BSC in Barcelona.
Further information on the workshop can be found on the event page.
Website of the ENES Portal:
https://portal.enes.org/