NEMO performance

To investigate the performance of the NEMO code we run the code using different numbers of processors and different grid configurations. We also investigate whether the particular compiler used makes any difference. The performance of NEMO is measured by examining the time.step file output by the code. This reports the model step number and time taken in seconds. The benchmark dataset runs for a total 60 model time steps and so our benchmark data reports the time taken for 60 time steps. All the the results presented in this section correspond to version 2.3 of NEMO. Results from version 3.0 of NEMO are presented separately.

The number of processors on which NEMO will be run is set in the source file named, par_oce.F90. This file also sets up the grid dimensions over which the model will be decomposed. The variables jpni, jpnj, and jpnij specify respectively the number of processors in the $i$ direction, the number of processors in the $j$ direction and the total number of processors. E.g. a 16x16 processor grid which runs on 256 processors would have jpni = 16, jpnj = 16 and jpnij = 256.