The workshop on Scalability took place at ECMWF on 14 and 15 April 2014.
Description
The challenges of highly parallel computing in areas such as observational data processing, data assimilation, the formulation of the model dynamical core and model output management have led to the formulation of a Scalability Project at ECMWF. The project will coordinate resources defining the future forecasting system across all scales. This will be in preparation for future high-performance computing architectures aiming at accurate, efficient and scalable algorithms and code structures.
The Scalability Project will rely heavily on external partnerships with numerical weather prediction centres, high-performance computing centres, academia and hardware providers. To prepare this collaboration ECMWF organised a 2-day workshop on 14 and 15 April 2014 in Reading. The workshop included presentations covering weather and climate science applications at scale, as well as numerical algorithm and hardware aspects towards exa-scale high-performance computing.
The workshop was co-funded by CliMathNet.
Objectives
The main objectives of the workshop were to define objectives for external collaboration by:
- identifying common areas of fundamental research towards exa-scale scalability of numerical algorithms, software infrastructures and code adaptation;
- exploring the potential for common/shared code components;
- defining future benchmarking strategies;
- facilitating intercomparison projects of different model components on selected architectures.
The workshop also aimed to define opportunities for funding/supporting the above activities by:
- exploring options for consortia targeting the European Commission Horizon 2020 framework;
- joining existing partnerships in science and industry;
- finding opportunities for access to external high-performance computing resources, including novel architectures.
Programme
Presentations
|
Welcome Erland Källén (ECMWF) |
 |
|
Introduction Peter Bauer (ECMWF) |
|
|
The ECMWF forecast model, quo vadis? Nils Wedi (ECMWF) |
|
|
Scalability and data assimilation at ECMWF Yannick Trémolet (ECMWF) |
|
|
Gung Ho and LFRic: Replacing the Met Office Unified Model Steve Mullerworth (Met Office) |
|
|
The ICOsahedral Nonhydrostatic (ICON) modelling framework: key aspects for computational efficiency and scalability Günther Zängl (DWD) |
|
|
Recent performance of NICAM on the K-computer and activities towards post-petascale computing Hisashi Yashiro (Riken/AICS) |
|
|
Implementation of COSMO on accelerators Oliver Fuhrer (MeteoSwiss) |
|
|
Running operational Canadian NWP models and assimilation systems on next-generation supercomputers Mark Buehner, Michel Desgagne, A Qaddouri et al (Env.Canada) |
|
|
Improving the efficiency of Arome-Arpege, achievements and plans Philippe Marguinaud, Météo-France |
|
|
Some views on the status and evolution of the Arome-Arpege dynamical core Fabrice Voitus (Météo-France) |
|
|
Streamlining HPC scenarios for future NWP Sami Saarinen (CSC) |
|
|
Running HARMONIE on Xeon Phi coprocessors Enda O’Brien (ICHEC) |
|
|
NEMO scalability issues Eric Maisonnave (CERFACS) |
|
|
Refactoring for the Xeon Phi Jacob Weismann Poulsen (DMI) |
|
|
The new Met Office NERC cloud model Michele Weiland (EPCC) |
|
|
The IBM Blue Gene/Q: Application performance, scalability and optimisation Mike Ashworth (STFC) |
|
|
ECMWF's future challenges in handling and manipulating Tiago Quintino (ECMWF) |
|
|
Data handling on the path to exascale Bryan Lawrence (NCAS) |
|
|
HPC for climate models: lessons from IS-ENES projects Sylvie Joussaume (IPSL) |
|
|
Scalability Bottlenecks towards Extreme Scaling of ICON Panagiotis Adamidis (DKRZ) |
|
|
PRACE resources and services toward climate and weather community Sergio Bernardi (CINECA) |
|
Working groups
| Report | |
| Working group 1 | |
| Working group 2 | |
| Working group 3 | |