Next: Running
Up: Parallel job
Previous: Inheritance
There are two required parameters and several optional parameters
to initialize a parallel job.
The first required parameter is (sect_pars) a Python dictionary that links
the jobidmentioned before, with a parameter object SEP.par.sep_pars.
This class knows how to read, write, add, and return a list of parameters.
An example of where a
job might need different parameters is if you are creating
an output cube that scans over a parameter space.
The second required parameter is a dictionary of parallel files that
are needed for the job. A dictionary
rather than a list is used to enable a parallel
job to be used in inversion. When inverting
the model and data vectors will change
depending on the inversion process. key in the dictionary enables
an easy way to locate (and possibly change Clapp (2004b))
the parallel file. The value in the dictionary is a parallel file
object discussed below.
Below is a list of some of the useful optional parameters:
- program
- The serial code that you want to run on each node. This is an
optional, rather than required parameter, because it is possible, and
in some cases useful, to extend the SEP.parjob.parjob class's
job creation mechanism.
- restart
- Whether or not we are restarting a job.
When restarting the status file will be read instead of created.
- global_pars
- A list of of parameters to run with every job,
again a SEP.par.sep_par object.
- attempts
- The number of attempts (defaulted to 2) to attempt
to start a job (on different nodes) before failing.
- restart_com
- A parameter that needs to be added to the calling
sequence when restarting a job.
- mfile
- The machine file to read the available nodes from (in the same
status convention as an MPI machine file. Defaults to a file name mfile.
- njobs
- The maximum number of jobs to run simultaneously. Defaults
to the number of processors described in mfile.
- verb
- The level of verbosity.
The initialization creates a machine class object (also
inherited from the status class). This object keeps
track of the available machines. An obvious extension
of this class is discussed in the future work section.
Next: Running
Up: Parallel job
Previous: Inheritance
Stanford Exploration Project
10/23/2004