EULAG Model for Multiscale Flows - Towards the Petascale Generation of Mesoscale Numerical Weather Prediction

摘要

EULAG is an established, highly parallel model for simulating fluid flows across a wide range of scales. It is known to scale well up to 16000 processors on IBM Blue Gene/W. It is noteworthy for its non-oscillatory integration algorithm MPDATA, advanced elliptic solver and generalized coordinate formulation. In this paper we focus on complex orographic flows and present the perspective of implementing EULAG as a high resolution weather prediction tool for Europe. As resolution of numerical models improves, numerical weather prediction enters the phase where traditional convection parameterization becomes obsolete and is replaced with a cloud-resolving approach. The boundary conditions, especially the topography, are becoming more and more complicated, demanding higher accuracy and better conservation properties from the numerical model construction. This calls for seeking fast and precise fluid solvers. We present preliminary results of simulations of the flow over realistic topography of the Alps, which proves the model capability to handle steep slopes. We demonstrate performance of the code on IBM Blue Gene/L architecture and compare different I/O strategies, from simple one-node operations to MPI I/O solution. An example of application of VAPOR, a tool for visualization and analysis of tera-scale sized data sets is provided.