闭排队网络基于并行仿真的灵敏度估计和优化算法

Sensitivity estimates and optimization algorithms based on parallel simulation for a class of closed queuing networks

基于 Markov性能势理论 ,对一类闭排队网络的灵敏度估计和优化 ,建立了一种行之有效的并行仿真算法。采用公共随机数 ,使所有的处理器使用相同的样本轨道 ,以减少各个处理器之间的通讯时间。在一台 SPMD并行计算机上的仿真实例表明 。

Based on Markov performance potential, an efficient parallel simulation algorithm is presented for sensitivity estimates and optimization of a class of closed queuing networks. The Common Random Number is applied to make all processors generate the same sample path, which removes the large broadcasting cost at the price of only adding a little workload. The simulation experiments on an SPMD parallel computer show that these algorithms can achieve nearly linear speedup for optimization of a class of closed queuing networks.