MMM SEMINAR NCAR
Large Atmospheric
Computation on the Earth Simulator (LACES):
an update on the project and on the MC2 model.
Michel Desgagne
Environment Canada
A long-term collaborative effort between scientists from the Earth Simulator Center (ESC), McGill University and RPN is now focusing the full lifecycle of hurricane Earl (September 1998) with the Canadian MC2 Community Model. The goal is to produce a 1 km horizontal resolution forecast over a large domain, which covers the tropical and extra-tropical re-development ofEarl. In fact, we wish to produce 8-9 days simulation on a fine-resolution domain of size 11000 x 8640 x 67. This reference simulation will be used to validate various lower resolution simulations and to improve our understanding of the extra-tropical transitions of hurricanes. Of course this represents a major computational effort that can only be performed – at least for the moment - on the Earth Simulator vector super-computer in Yokohama (Japan) using more than 90% of the total resources of the system for up to 10 full days of computation. This project also poses a real challenge for the MC2 modelling system itself considering that the computing work is distributed on 3960 vector processors (495 out of 540 nodes). Inter-processor communications are potentially becoming an issue with such a large number of processors. It also represents a real test for the numerics of the elliptic solver considering that an order 10 Giga-equations are solved at once by an iterative 3D pressure solver based on FGMRES. I/O is also a very important issue, which must be treated carefully. Even the simple display of meteorological fields on horizontal planes is far beyond the limit of current display technology. Project LACES is now into its production phase. Results from the tropical phase of the simulation can already be used for various studies. Technical aspects of the setup required to achieve such a large simulation will be described. Machine performance will also be presented as well as preliminary meteorological results. Finally recent developments on the MC2 model - in particular those related to the elaboration of a unified dynamical kernel for atmospheric and oceanic flows - will be discussed.
Thursday, 17 February 2005,
3:30 PM
Refreshments 3:15 PM
NCAR-Foothills Laboratory
3450 Mitchell Lane
Bldg 2 Auditorium
(Rm1022)