论文部分内容阅读
对Ｍａｒｋｏｖ性能势理论在一类闭排队网络中的应用给出了一种高效的并行仿真算法．针对仿真中占总运算量７０％以上的一类参数矩阵运算的特点，提出了一种新的矩阵元素对处理器的分配方法：“螺旋式”划分法，使该部分矩阵运算在所有处理器上达到了完全负载平衡；针对仿真中样本轨道数据的大量的广播通讯开销，该文提出用“公共随机数”使所有处理器产生相同的样本轨道，从而以增加很小工作量的代价去掉了这部分通讯开销．另外，由于仿真中占总运算量９０％以上的两部分矩阵计算的特征：没有矩阵元素在各处理器间相互交换的通讯开销，该文算法能广泛地适用于各种体系结构的并行机．最后给出了该并行算法在一台ＳＰＭＤ并行机上的计算结果，显示出该算法能取得接近线性的加速比和较高的并行效率．
An efficient parallel simulation algorithm is given for the application of Markov potential theory in a class of closed queuing networks. Aiming at the characteristics of a kind of parameter matrices operation which accounts for more than 70% of the total computation in simulation, a new method of allocating matrix elements to processor is proposed: “Spiral” method, which makes the part of matrices operate on all processors To achieve a complete load balancing; Simulation of sample orbit data for a large number of broadcast communications overhead, the paper proposed “public random number” so that all processors produce the same sample orbit, which in order to increase the cost of a small workload removed This part of the communication overhead. In addition, due to the computing characteristics of the two-part matrix which accounts for more than 90% of the total computation in the simulation, there is no communication overhead for the matrix elements to exchange among the processors. The algorithm can be widely applied to parallel machines of various architectures. Finally, the calculation results of the parallel algorithm on a SPMD parallel machine are given, which show that the algorithm can achieve near-linear speedup and high parallel efficiency.