An embarassingly parallel framework for running EGSnrc/BEAMnrc/DOSXYZnrc, FLUKA, MCNP/MCNPX, GEANT4 and PENELOPE on grid and cluster computers

We present a computing framework for Monte Carlo simulation of radiation transport. In conformity with our belief that Monte Carlo studies should be anchored to physics but not to any particular code, we have enabled different Monte Carlo codes in this environment. The framework embraces the EGSnrc family, FLUKA, MCNP/MCNPX, GEANT4 and PENELOPE; dedicated and non-dedicated resources; shared and non-shared filesystems; local clusters and nationsl grids; Unix/Linux family as well as Windows operating systems. The framework works whether or not the executing node has been pre-installed with the Monte Carlo code. It works with resource brokers such as Condor, Nimrod, PBS, SGE, OMII GridSAM and gLite. Validation and quality assurance become very important when using heterogeneous computers beyond local administration-particularly if simulation results are used for making decisions which potentially save or harm human life e.g. patients seeking medical treatment or staff exposed to radiation hazards. We describe the steps we take before and/or after simulation execution to gain confidence in simulation results. Before execution of the full simulation, smaller test jobs may be run so that results may be validated against known values. This exercise is similar to the test suites available in MCNP/MCNPX, sometimes replaced by or coupled with problem-specific tests designed by the user. After execution of the simulation, results from different batches of radiation histories are compared for statistical agreement. A series of illustrative examples of the use of this computational framework will be presented, involving inter-code comparisons for medical physics applications.