3.1 Job control with mrun

The start of model runs in principle takes place via executing the shell script mrun. With this script it is possible to operate the model both interactively and in batch mode. Batch operation is possible on the local computer as well as on a remote computer (supposed that a batch queuing system is available). Here only the batch operation on a remote computer is described, because this represents the usual mode of operation due to the large computing time requirements of the model. The interactive execution differs only in some points and is described in chapter 3.6.

By the call of mrun a complete batch job is produced, transferred to the user determined remote computer and submitted to its queuing system (LoadLeveler, PBS, etc.). After processing of the job, the so-called job protocol is sent back to the local computer of the user (the local computer is the computer, on which the user calls mrun). The following actions are implemented by the job, amongst other:

From this list it becomes clear that two of the substantial functions of mrun are the supply of input files for the model and the storage of output files produced by the model. The model exclusively works with so-called local files, which lie in the temporary working directory created by mrun. The “local” names of these files are determined by the model (in the appropriate OPEN instructions). In chapter 3.4 all possible input and output filenames are specified. In contrast to this, the input files, made available by the user, usually lie in permanent directories (e.g. in the /home directory) and the output data, produced by the model, may also be stored there. The directory names are arbitrary. The actual file name (without path) of these input and output files is specified by an option of mrun. All input and output files of the model receive this file name (e.g. abcde), but each individual file is still provided with an extension (e.g. abcde_par, etc. abcde_dat) - otherwise the files would not be distinguishable. The linkage of local file names with the names of the permanent directories and the file name extensions takes place in a configuration file, which mrun expects to find in the respective current working directory of the user. A detailed description of this configuration file - which still offers a variety of further functions – is found in the documentation of the shell script mrun (sorry: this is only available in German, so far. If you think that you would profit from an English version, please give a message to the PALM group). In the next chapter a simple example of a minimum configuration file is described.

A simple call of mrun could read as follows:

The model is started by option -h ibmh on the IBM Regatta hanni at the HLRN, i.e. on the local computer a batch job is generated, transferred to the remote computer (IBM) and submitted to a suitable queue of the queuing system. With the option -d the so-called basis file name (abcde) is specified, from which, together with the path names and file name extensions, the complete file names (including the path) of the permanent input and output files are formed. Usually the specification of a set of further options is necessary, e.g. the CPU time needed by the model and the memory size needed as well as, on parallel computers, the number of processing elements which are to be used (options -t, -m and -X). A precondition for the call of mrun of course is that the installation of the model on the local computer and the remote machine is complete (see chapter 5.0).

Beyond these options a variety of further options for model steering can be used in the call of mrun. These are fully described in the mrun documentation (in German).

Last change:  14/04/05 (SR)