Load Balancing in Highly Parallel Processing of Monte Carlo Code for Particle Transport.
暂无分享,去创建一个
In parallel processing of Monte Carlo (MC) codes for neutron, photon and electron transport problems, particle histories are assigned to processors making use of independency of the calculation for each particle. Although we can easily parallelize main part of a MC code by this method, it is necessary and practically difficult to optimize the code concerning load balancing in order to attain high speedup ratio in highly parallel processing. In fact, the speedup ratio in the case of 128 processors remains in nearly one hundred times when using the test bed for the performance evaluation. Through the parallel processing of the MCNP code, which is widely used in the nuclear field, it is shown that it is difficult to attain high performance by static load balancing in especially neutron transport problems, and a load balancing method, which dynamically changes the number of assigned particles minimizing the sum of the computational and communication costs, overcomes the difficulty, resulting in nearly fifteen percentage of reduction for execution time.
[1] Kiyoshi Asai,et al. Difficulties in vector-parallel processing of Monte Carlo codes , 1997 .
[2] Forrest B. Brown,et al. Monte Carlo methods for radiation transport analysis on vector computers , 1984 .
[3] Kiyoshi Asai,et al. Development of Monte Carlo machine for particle transport problem , 1995 .
[4] E. S. Troubetzkoy,et al. Monte Carlo radiation penetration calculations on a parallel computer , 1973 .
[5] Kiyoshi Asai,et al. Vectorization of the KENO-IV Code , 1986 .