iCAS attendees share advances and challenges in modeling HPC hardware, workflows and data
More than 60 representatives from weather and climate computing centers around the world discussed advances in modeling and HPC systems as well as emerging challenges in data analytics, workflows, and data management facing the community at the 14th biennial International Computing for the Atmospheric Sciences (iCAS) Symposium in Annecy, France, September 10-14, 2017.
“We’re in the midst of a transformative era for computing in the atmospheric sciences,” said Anke Kamrath, director of NCAR’s Computational and Information Systems Laboratory, which hosts the symposium. “The next-generation of computing hardware has a number of possible futures, models are capable of ever higher resolution, analysis workflows have to cope with torrents of model output, and machine learning and data science are poised to upend how we think about all phases of modeling and analysis.”
The Tuesday keynote from NCAR’s Andrew Gettelman underscored the modeling challenges faced by science teams at NCAR as the push NCAR climate and weather models into uncharted territory. Attendees also heard from Young C. Kwon of the Korea Institute of Atmospheric Prediction Systems (KIAPS) on the modeling challenges faced by the next-generation global model at KIAPS, and James Kinter of the Center for Ocean-Land-Atmosphere (COLA) Studies at George Mason University highlighted the scientific issues and tradeoffs that require balancing between ensemble size and higher model resolutions.
“Increasing resolution is not a panacea for climate prediction,” Kinter said, noting that higher resolution simulations may require larger ensemble sizes, not smaller.
On the hardware front, Paul Messina of Argonne National Laboratory in his Wednesday keynote discussed the goals, plans, and challenges for the U.S. Department of Energy’s Exascale Computing Project. “The reason for doing exascale is not the new hardware, but the applications,” Messina said, in describing the efforts to prepare models for future exascale systems. Haohuan Fu of China’s Tsinghua University described similar efforts to scale and optimize climate and weather models, including several Gordon Bell prize-winning successes, for the Sunway TaihuLight system, currently the number one system on the Top500 list.
NCAR’s Rich Loft summarized the “Cambrian explosion” in the evolution of HPC hardware in his Project Zeta presentation. Attendees also heard presentations on the latest developments from leading vendors and iCAS sponsors, including Cray, Hewlett Packard Enterprise, IBM, Intel, Mellanox, and NVIDIA.
Several presentations shared information on facilities, computer procurements, and other developments at Japan Agency for Marine-Earth Science and Technology (JAMSTEC); the European Center for Mid-range Weather Forecasting (ECWWF); the U.S. National Energy Scientific Computing Center (NERSC); Korea Institute of Science and Technology Information (KISTI); Shared Services Canada; NASA Center for Climate Simulations; the Deutsches Klimarechenzentrum (DKRZ); and the Australian Bureau of Meteorology.
A wide range of presentations focused on challenges and new developments around data— input from novel observational platforms (Pete Beckman, ANL, “Waggle and the Array of Things”), assimilation into forecast systems (Glen Romine, NCAR, and Ibrahim Hoteit, KAUST), workflows for data analytics (Ilkay Altintas, San Diego Supercomputer Center), and applications for data science and machine learning.
The agenda included two panels on specific data topics—data compression and CMIP6 analytics and workflows. On data compression, panelists emphasized the issues and benefits for scientists in adopting lossy compression as a way to mitigate the challenges in storing, managing and accessing growing volumes of modeling output. The CMIP6 panel discussed efforts across Europe to prepare analysis platforms and workflows in anticipation of community-wide interest in the CMIP6 modeling output planned from dozens of modeling centers worldwide.
The full program, along with speaker bios and available presentations, can be found on the iCAS 2017 web site. We look forward to seeing new and returning attendees at the next iCAS Symposium in 2019.