Optimum Routing and Forwarding Arrangement in Bufferless Data Vortex Switch Network

Abstract

Data Vortex networks with different routing and forwarding path arrangements are studied. To keep the bufferless routing nodes, the binary tree decoding stages in Data Vortex network are shown to achieve best combined routing performance in throughput and latency. When k is too large, the general k-ary decoding stages introduce too much deflection latency penalty, so this paper focuses on the performance comparison between the binary tree (k=2) decoding and 4-ary decoding (k=4) cases. The results have shown that even though a much smaller number of cylinder levels are required in a network of k=4, there is only slight reduction in the average packet latency within the network. On the other hand, the traffic backpressure induced throughput limit and the additional implementation cost in 4-ary decoding networks have proved that binary tree decoding stages offer the optimum performance under bufferless operations. The study has in part shown the advantages of Data Vortex network topology over many two dimensional multiple stage interconnection networks (MIN) if only deflection is allowed for bufferless operation.