A load balancing strategy for the radiation transport code event

Uneven assignment of the computational load among processors causes a significant portion of the total solution time being spent on processor synchronization. To address this inefficiency, we propose two different guiding parameters: one for determining whether the computational load is evenly distributed among the processors, and another for determining whether improvement in the speed up can be achieved. These two parameters are computed from a fast, auxiliary diffusion calculation, and, based on their values, a load balancing technique is applied to repartition the mesh. This strategy has been implemented in the finite element-spherical harmonics radiation transport code EVENT and the speed up achieved for higher-order angular approximations before and after load redistribution is compared for the three dimensional C5G7Mox reactor problem. A significant improvement on speed up is achieved with higher number processors after load redistribution. (authors)