Data Structure and Memory Management in the NCAR/LLNL (National Center for Atmospheric Research/Lawrence Livermore National Laboratory) Community Climate Model (CCM1L)

摘要

Extensive modifications of the Community Climate Model Version 1 (CCM1) were required, after this computer code was imported from the National Center for Atmospheric Research, to enable efficient execution in the computing environment of the Lawrence Livermore National Laboratory. Most of the modifications involved storage and transfer of large data sets as the model executes. The management of these data sets was generalized to permit users to select among a variety of fast storage media-internal memory, solid state storage devices, or disk-depending on their availability and their suitability for optimum performance on present and future supercomputer systems. We describe these additions to the CCM1 code and related modifications to its data files. In the optimum configuration of the LLNL version of CCM1, the code is run on a Cray XMP computer with attached Solid-State Storage Device; working files are maintained in internal memory or the SSD, while input and output files are buffered from and to disk. This procedure results in a great reduction in input/output overhead. 2 refs., 19 figs. (ERA citation 14:009146)