基于多变量判决函数的最优路由策略

Optimal Routing Policy Based on Multivariable Decision Function

针对分布式并行处理系统中路由算法数据包的路由选择问题,提出一种改进的最优化路由策略。从输入数据包得到数据包前后到达时间分布Pt(x)和包大小分布Pp(x),采用权值函数通过对平均前后到达时间、平均包大小和向量的不断学习获得所有包的最小化平均延迟。仿真结果表明,该策略不仅在处理器数量发生变化,而且在包前后到达时间分布和包大小分布改变的情况下,都能获得所有包的最小平均延迟。

Aiming at the disadvantage of existing routing algorithm in distributed parallel processing system,a novel and effective routing policy is proposed. The concrete implement is to obtain the packet fore-and-aft arriving time distribution Pt（x）and the packet size distribution Pp（x）from input packets,and a weight function is introduced to achieve the minimal average delay of all packets by learning continuously average fore-and-aft arriving time,average packet size and vector. Simulation result shows that the minimal average delay of all packets can be obtained not only in the condition of the varied number of processors,but also in the condition of the changed packet fore-and-aft arriving time and the packet size distribution.