Monte Carlo modeling of linear accelerator using distributed computing.

PURPOSE The distributed computing implementation of the EGSnrc Monte Carlo system using a computer cluster is investigated and tested. METHODS The computational performance was tested for various scenarios with different number of computers used in order to assess the efficiency of the cluster. The presented computation times and efficiencies include the linac head modeling with full simulation of the multi leaf collimator (MLC) geometry (including tongue and groove) and stereotactic radiosurgery cones as well as the radiation transport simulation and dose computation within water phantom and patient geometry. RESULTS The simulations performed in the cluster environment had the same total number of histories recorded and simulated as the simulations in a single computer. The statistical uncertainty achieved was the same for all scenarios. CONCLUSIONS The investigated approach shows almost linear performance scaling vs number of computers involved.