Received scripts from Camiel Severijns to postprocess data from a CMPI5 experiment.
The spectral files (in GRIB format) should be splitted into files per variable and
afterwards the scripts should generate yearly NetCDF files. The postprocessing of
1 year of data takes several hours. If archiving is included, it takes 6 hours.
- end of May 2011: initial report on benchmarks of CDO on Power6 and comparison
to experiences from SGI system at KNMI.
- end of July 2011: more performance results from Intel-based systems.
-end of September 2011: suggestions for improvements of CDO, possibly related to
the fact that compiler optimizations can dramatically increase performance. Maybe
arrays are not consecutively accessed?