Climate Modeling using High-Performance Computing

Mirin, A A ; Wickett, M E ; Duffy, P B ; Rotman, D A

Lawrence Livermore National Laboratory

关键词: Lawrence Livermore National Laboratory;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;    Physics;    Computers;    Simulation;   

DOI  :  10.2172/15011612
RP-ID  :  UCRL-TR-210224
RP-ID  :  W-7405-ENG-48
RP-ID  :  15011612

美国|英语

来源: UNT Digital Library

【 摘 要 】

The Center for Applied Scientific Computing (CASC) and the LLNL Atmospheric Science Division (ASD) are working together to improve predictions of future climate by applying the best available computational methods and computer resources to this problem. Over the last decade, researchers at the Lawrence Livermore National Laboratory (LLNL) have developed a number of climate models that provide state-of-the-art simulations on a wide variety of massively parallel computers. We are now developing and applying a second generation of high-performance climate models. As part of LLNL's participation in DOE's Scientific Discovery through Advanced Computing (SciDAC) program, members of CASC and ASD are collaborating with other DOE labs and NCAR in the development of a comprehensive, next-generation global climate model. This model incorporates the most current physics and numerics and capably exploits the latest massively parallel computers. One of LLNL's roles in this collaboration is the scalable parallelization of NASA's finite-volume atmospheric dynamical core. We have implemented multiple two-dimensional domain decompositions, where the different decompositions are connected by high-speed transposes. Additional performance is obtained through shared memory parallelization constructs and one-sided interprocess communication. The finite-volume dynamical core is particularly important to atmospheric chemistry simulations, where LLNL has a leading role.