How Abed worksΒΆ
Running the tasks on the cluster is done as follows:
In the PBS the following preprocessing steps are taken:
A
results
directory is created on the scratch filesystem (see above)An email is sent to the current user with a brief summary of the number of tasks to be performed. This is sent to the users email, an may require a forwarding file to be delivered correctly.
Optional modules are loaded
Optional environment variables are exported
mpicopy
is used to copy files from the home directory to the scratch filesystem
abed run
is executed throughmpiexec
. This requires Abed to be in the PATH variable. This command is run in a timeout command, with the execution time reduced by a configurable number of seconds. This should allow enough time for the post-processing stepsThe postprocessing steps are:
Create a
bzips
directory in the current release directoryCreate a compressed archive for each dataset in the result directory, this is done in parallel through
pbzip2
Copy the compressed files to the current release directory
Sent an email to the user that the task is completed.