You are here: Home » Archive » Phase 2 » Project » NA2 (WP3): Towards next generation models

NA2 (WP3): Towards next generation models

 

Objectives


The main objective of this work package is to prepare the climate community for exascale and better connect those working on climate models by establishing some code convergence and shared understanding of divergent codes. We will promulgate best practice model development to support the running of these codes on the massively parallel, unreliable, heterogeneous machines we expect in the exascale future. Two broad areas will be addressed: code convergence and programmability.

In terms of code convergence, we have two tasks:

1. Establishing common radiation tools, and

2. Convergence: documenting existing models in terms of performance, portability issues and programming models in such a way that future convergence activities can be planned.

 

In terms of programmability, we have one task:

3. Technology Tracking: tracking the hardware/software environment within which climate and Earth system models will be executed, taking into account the new foreseen exascale computing systems and related issues (new parallel approaches and new coupling framework strategies).

Community building with common radiation tools

We begin community building with radiation code, since it is both fundamental to the climate problem and

it is a computationally expensive part of climate models. Radiation is handled in atmospheric models by

parameterisations that use the spectroscopic properties of gases to carry out clear sky radiative transfer

alongside with scattering processes which result from the presence of particles and/or clouds. In general the

clear sky radiative transfer is a well understood problem around which consensus can be built – while the

detailed spectroscopic behaviour of the gases and the scattering processes are both under active research.

We aim to build a community around a common European approach to this by working together to develop a radiation library providing radiation solvers so as to split out the research/model specific elements around the presence/structure of particles and spectroscopic behaviour from the relatively well understood radiative transfer problem.

A key part of evaluating the success of this approach in terms of both the models and the radiative codes within them, will be the comparison of models with observations. This comparison is facilitated by the use of simulators such as those in the Cloud Observing System Package (COSP). We will incite the European climate community to advance the use of these packages.

Developing Convergent Model Codes

Moving on to further convergence will require the comprehensive documentation and analysis of several

models so modelling groups can be better informed of what each other are doing at a detailed level. Knowledge of both scientific properties and performance will be required. Key code components will be identified, and performance assessed by exploiting code simulators (which allow one to assess how efficiently software will run on future architectures), in particular the BSC DIMEMAS simulator. Work on scaling and portability will also involve different High Performance Computing environments, and exploit work carried out in IS-ENES (e.g. on parallel input/output). The portability of the libraries on several architectures will be also investigated (which will direct link with the work on technology tracking discussed below). Establishing across Europe best practice on programmability and the usage of code performance simulation will be crucial for effective progress. A range of mechanisms will be used both to communicate within the climate community and exchange information and requirements with the wider community.

Technology Tracking

Clearly it will also be important to look out at what the wider exascale software community is doing, and the relevant hardware trends, and bring that information into the climate community. In particular we need to bring together experts from the two communities to document the issues regarding the upcoming computational architectures, and to look at opportunities from better exploiting parallel numerical libraries. We also need to feed the needs of the climate community into the wider HPC community. Examination of existing solvers will be necessary to support this “inter-disciplinary” communication. The main issues that will be addressed include: memory management and data transfer between the Graphic Processing Units and the Central Processing Units; the portability across different accelerator technologies; and the programming languages (OpenCL, HMPP, Cuda-Fortran ...) used or usable. The main exascale issues (data locality, low memory/core, exploitation of massive multithreading, new languages paradigms, etc) will be taken into account. New coupling strategies will also be evaluated since they have a strong impact on the code structure and performance of climate models.


Deliverables

D3.1 Report on the technology tracking

This deliverable will give a look out at other European and global activities addressing exascale issues, and bring that knowledge into the European climate community. The relevant topics covered by the report are: description of the exascale technologies; an overview of the new parallel approaches for the climate models; and strategies for coupling.
D3.2 Report on strategies for developing convergent model codes
D3.3 Report on common radiation tools



 Title   Type 
NA2 Working documents Folder
© Copyright ENES Portal 2011

flag_eu.png

 

IS-ENES3 has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824084