ECHAM5 input/output

Input and Output files

ECHAM5 requires several files for starting the integration. These file contain information for the description of the initial or re-start state of the atmosphere, for the description of assumed conditions during the integration, e.g. sea surface temperature, or for the initialization of parameterizations. Resulting data are written to one or more data files, and the final state of the integration is stored in a re-start file that may be used to continue the experiment later.

File formats

The file formats are NetCDF for the description of initial or re-start states. Resulting data are stored in NetCDF or GRIB, the latter is a common standard for meteorological data and was default for ECHAM4 and older cycles. ASCII or readable text format is used exceptionally for some small initialization files.

Data streams

ECHAM5 is designed to read/write from/to more than one file. This is specifically useful when ECHAM5 is hosting so called "sub models" which require their own input/output, or when extra fields are should not be mingled with the standard output fields. Therefore data i/o is organized in terms of "data streams".

Input and Output in a parallelized world

ECHAM5 is designed to run on a parallel computer architecture. Input and output in such an environment usually is a bottle neck and needs special care. In ECHAM5, I/O is concentrated on one specific processor, the io-processor. This is usually faster than parallel I/O because data transfer from a non-io-processors to the io-processor is more efficient than the interruption for I/O. In ECHAM5, interruption for I/O is done on the io-processor only. In practice this means, that before each I/O block one has to find out if the current work is done on the io-processor or on any other processor.