When

Sep 12, 2016 10:00 AM to Sep 13, 2016 07:00 PM
(Europe/Berlin / UTC200)

Where

Rockville (USA)

Attendees

George Mozdzynski (ECMWF)

Add event to calendar

iCal

Presentation by George Mozdzynski (ECMWF) . Plenary Speaker, on “Addressing Future Scalability and Power Challenges at the European Centre for Medium-Range Weather Forecasts (ECMWF)”.

Abstract

ECMWF is pursuing a long term Programme on Scalability that aims at developing the next-generation forecasting system addressing the challenges of future exascale high-performance computing and data management architectures. The Programme is required to optimize system performance allowing ECMWF to fulfil its strategy within expected funding and environmental constraints. Key components of model, data assimilation, code adaptation to new and emerging hardware solutions as well as data pre- and post-processing are included. The Scalability Programme is expected to produce new capabilities at ECMWF, namely:

  • an integrated forecasting system combining a flexible framework for scientific choices to be made with maximum achievable parallelism;
  • portable code structures ensuring efficiency and code readability exploiting a range of expected future technologies;
  • metrics and framework for code testing allowing quantitative assessment of scalability.

In this talk I will present an update on the ECMWF Scalability Programme, and recent developments and experiences on running high resolution forecast models on today's petascale supercomputers.

About this event

Advancing X-cutting Ideas for Computational Climate Science (AXICCS 2016) is a workshop to discuss bold, new computational ideas to address longer-term science needs for climate modeling.

As the climate science community explores a host of critical science questions to be answered over the next 10+ years, with the aim of informing stakeholders about the ongoing changes in global and local climate, there is a growing recognition of the expanding requirements for multiscale, global, coupled Earth system models. We expect them to provide much more detail and fidelity, with a much better understanding of their uncertainties, while still executing robustly and efficiently on ever larger and more complex computing systems.