DOCS : MONTE CARLO

Monte Carlo workflows

justIN relies on matching unprocessed input files with wrapper jobs which start on worker nodes at sites. To be able to include Monte Carlo workflows in the system, it is necessary to create virtual counter files in the justIN database's list of files. These counter files are allocated to jobs one by one to keep track of the number of Monte Carlo jobs which are required. This framework can be used for other types of workflow which also do not require real input files from bulk storage, such as such parameter scanning applications.

The justin command man page explains how to submit workflows in this case. In short, the MQL expression is replaced by "monte-carlo NNNNNN" where NNNNNN is the number of jobs required. The justin command has an option --monte-carlo which does this.

When the Finder agent processes a Monte Carlo workflow, it creates the correct number of virtual counter files in the database, each with a name of the form "monte-carlo-RRRRRR-XXXXXX" where RRRRRRR is the Workflow ID number assigned by justIN and XXXXXX is a number between 000001 and NNNNNN. Each file has a single replica on the virtual RSE MONTE-CARLO, with PFN consisting just of its XXXXXX number. These file and replica names are readily accessible within the jobscripts and so can be used to derive unique names for the output files, which may be useful for debugging purposes.