Navigation and service


Climate simulation

Climate simulation

Understanding the evolution and changes of global climate is of utmost importance in the 21st century. The complexity of climate simulation is reflected in the structure of codes in the field. In this case, the application, called EMAC, consists of two coupled models. The atmospheric model represents pressures, currents, temperatures and related magnitudes of Earth's atmosphere. Coupled to this base model, a chemical simulation package analyses fine grain interactions between chemical elements.

The atmospheric model requires a significant number of transformations and data transpositions, resulting in constant global communication and lack of overall scalability. On top of that, photochemical effects caused by changes in sun light over the Earth result in a very significant load imbalance and therefore worsen an already suboptimal scalability. The processing of these local photochemical effects consumes most of the time in these simulations, due to the synchronicity of the model and its heavy computation requirements. In DEEP, individual tasks of the chemical model are offloaded to the Booster dynamically, effectively reducing the load imbalance and allowing the code to scale further than before, due to two reasons: 1) The atmospheric model can be kept as small as possible to avoid excessive communication, without hindering the heavy computing parts of the code; and 2) the load imbalance, the main concern to scale the code, is effectively eliminated. These benefits give an extra edge in scalability and performance for the EMAC community.

These simulations are done by The Cyprus Institute (CYI), Cyprus.

 

"Studying the chemistry and physics of the atmosphere we employ a complex Earth-system simulation coupling a general circulation model with local physical and chemical processes. Running on parallel supercomputers the global processes have high communication demands while the local processes are inherently independent. The DEEP architecture is naturally suited to these tasks with global components running on the Cluster nodes exploiting the high-speed Xeon processors and local components running on the highly-parallel Xeon Phi co-processors. However, different distributions of work cause different amounts of communication. By balancing communication versus computation the DEEP concept provides us with a new degree of freedom allowing us to distribute the different work elements at their optimal parallelisation." - Hendrik Merx, CYI


Servicemeu