实现虚拟输出队列调度的神经网络方法

A Neural Network Method Achieving Virtual Output Queuing Scheduling

摘要:

基于虚拟输出队列（VOQ）缓存的Crossbar交换结构,提出了一种Hopfield神经网络（HNN）控制的信元交换调度方法.通过选取合适的能量函数,并在其中采用一种新的队列优先级函数,实现了信元的高效交换控制.计算机模拟结果表明,该算法可以将吞吐率提高到0.998,信元丢失率大大降低,时延特性也有很大改善.

Abstract:

Based on the Crossbar switching fabrics with virtual output queuing(VOQ) buffers,an effective Hopfield neural network(HNN) based control approach for scheduling cell is proposed.By defining a new queuing priority function,in which priority is proportional to the number of cells in the buffer,and choosing appropriate parameters in energy function,the neural network scheduler can improve the performance of switching fabrics on quality of service(QoS) greatly.Simulation results show that the proposed approach can improve throughput to 0.998,the cell loss rate is decreased near to 0 and the cell delay is reduced.