基于渐进神经网络(GNN)的分组调度  

Applying neural network in input-queued switch

作　　者：路平[1] 王振家[1] 

机构地区：[1]空军工程大学工程学院,陕西西安710038

出　　处：《计算机工程与设计》2004年第1期58-60,共3页Computer Engineering and Design

摘　　要：现代通信网络中的大规模并行传输,意味着在交换机中的一个时间片内可能会有不止一个分组申请抵达同样的目的端,即引起了无冲突传输的问题。运用渐进神经网络(GNN)成功地实现了分组调度,并在网络的训练过程中设计了两个新的策略,即队首分组等待时间大于阈值无条件传输;队列分组数量达到缓冲器容量时队首分组无条件传输。仿真结果表明,运用神经网络调度,分组的丢失率为0,吞吐量达到了100％,而平均延迟也可以通过阈值设定进行控制。In modern communication networks , the parallel use of multiple channels means that one time slot more than one packet may have the same destination, which cause the problem of conflict-free traffic assignment. The problem of the conflict-free packet switch was successfully solved by the use of neural network. During the process of training, two new strategies were devised, the first strategy is once the waiting time of the first packet in any queue is greater than a threshold, the first packet must be transfered at once; the second strategy is once the number of packet in one queue is equal to the capacity of buffer, the first packet must be transfered at once. The result of the simulation suggest that by the use of neural network, the rate of lost packets is zeros, and the average delay can be controlled by the set of threshold.

关 键 词：渐进神经网络 GNN 分组调度 交换机 排队 通信网 

分 类 号：TN915.05[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]