Parallel Framework for Monte Carlo Simulations of Particle Tracking

MONTE Carlo (MC) simulations in physics are known to be highly compute intensive applications because they track the life of each simulated photon, which interacts multiple times with surrounding tissue, generating secondary particles. Therefore, there is great interest in accelerating MC simulations with technologies such as shared-memory multiprocessors, distributed-memory multicomputers and Graphics Processing Units (GPU). One of the most important examples in this respect is Geant4 and its parallelization efforts, namely ParGeant4 [1] and multithread-Geant4 [2]. Both cases implement event-level parallelism to partition the simulation among compute units. ParGeant4 employs a master-slave approach at the process level to send objects to be simulated in different computers. In multi threaded Geant4, multiple threads are created and each one simulate and track photons. In this case the simulation happens inside a multiprocessors, so there is no need to send objects across a network. Another important examples are Cluster Gate, ParGate [4], and Hybrid Gate [5], which are parallel versions of Gate. © 2015 IEEE.