The advantage of running a parallel code is that the work load can be distributed across more than one processor element (PE), typically between about 24 and 144 PEs. This should significantly reduce the time it takes a run to complete. As the data is shared across PEs, it should also reduce the size of the arrays needed on each PE for a given run, which allows runs to be made that would otherwise have exceeded the virtual memory limit, or even the total memory available.

The IGCM is a spectral global climate model, which has been run in the Meteorology at the University of Reading for many years. Versions 1, 2.2 and 3 of the IGCM have all been parallelised, and these web pages provide information on running them. Of the documentation that exists for the IGCM, most of it is contained in the folder that Mike Blackburn has put together. These pages cover the changes made to the IGCM in making it parallel, and provide information on running the parallel versions:

  • Main changes for Parallel IGCM: summarises the main changes that have been made to the IGCM to make it parallel.
  • Parallelisation of the IGCM: explanation of how the code has been parallelised
  • Nupdate directives: how nupdate directives have been used to create three parallel versions of the code in one file.
  • File names: for the parallel code, the input and output files have been given names. This page details the names given to the fort.* files from the non-parallel versions
  • Where to find code: summarises which directories contain the files that are needed to run the IGCM
  • Scripts for running IGCM: details some changes needed in the run scripts for the parallel code and lists where some example run scripts can be found.
  • Efficiency of IGCM: some timings for running the parallel version of the IGCM

The links on the left hand buttons also point to some of the other documentation available and some of my personal notes.

Things to do now


[an error occurred while processing this directive]