Open makeclean opened 11 years ago
Checked functionality of Bill Taylors script, running an simple mcnp geometry of a test problem originally used by Thad. The script split a the calculation into 30 serial calculations. We ensure that the RN strides do not overlap by changing the random number seed in each calculation in a manner that MCNP could not used enough random numbers to allow them to overlap.
This seems dangerous, instead for safety I suggest that seed is kept constant and instead
hist is changed. This ensures that the RN seed cannot be overlapped in a guaranteed safe
manner.
The model contains 8 volumes, several surface/cell tallies and 2 meshtal files. The meshes aren't large, containing less than 10^4 elements. However is a simple test for small scale models. Each calculation will run for 20 computer minutes before finishing. Since all simulations are built using a DiAG, when finishing up, scripts will be launched which combine the results to single mctal and meshtal files.
NOTE: Still running, will update when complete 15:19 08/04/2013
I agree with the above warning. For historical perspective, this idea of changing the seeds facilitates using fixed computer time in each job and not a fixed number of histories. It is questionable whether we want to support this in the near term or long term.
16:39 (08/04/2013) 24 of the 30 tasks are complete
Completed by 01:33 (09/04/2013), may have been on systems with particularly high loads
Was there evidence that any particular job took a long time? Or that it sat in the queue for a long time? Or that it was pre-empted and restarted multiple times?
The current working version can split a MCNP job into a number of chunks, the script currently checks for
The script then checks for the existence for a number of different files based on the run, damgc input if its required, tetmesh mesh if required, etc. The script will exit with errors if certain conditions are not met, such as nps being specified or prdmp being absent.
The script 'pre initialises' the runtpe files for the run, the runtpe files are unique to each run if tetmesh tallies are required, but otherwise are identical. The reason for the lack of homogeneity is down to the "out=" keyword, it was deemed safer to force the use of the "out=" keyword to produce uniquely named output tetmeshes to avoid clashes.
The script then generates a directed acyclic graph (DyAG) to specify the run hierarchy and monitor job completion. A DyAG graph is also built to produce the combined output.
This functionality exists as of commit a461ba1e7b98cdd9c8fee7f246f1de752950e99a, I am currently testing all of the features and will close the issue when testing is complete.
Need to check what has been done by Bill in order to determine MCNP functionality.