Performance Analysis Of 3d Noc Using Label Switching For Streaming Platform

Abstract

In this research work, a label switched predictor router is designed using Max flow algorithm in 2D Network on chip to reduce the power consumption in transmission of data packet among the communication entities. The power consumption in Network on Chip can be reduced by avoiding transmission of redundant data. So, a predictor is included in the design of 3x3 network for a 8 bit data size and 4 bit label size. The design is implemented in Xilinx software and ISIM simulator where, the power consumption in label switched router is 177.14mW and in label switched-predictor router is 155.85mW. There is 12 % improvement in power consumption. When the size of network increases; the power consumption for the repeated data inputs will be reduced by this implementation. A Label Switched Router (LSR) is implemented for a 2D and 3D NoC where Label Switching Technique using Dimension order routing algorithm. Here 3 x 3 NoC is used and extended as two cascaded 3x3 NoCs and compared its performance with 3D NoC of 3x3x2 network structure. In cascaded network, based on the nearest destination node, Latency is 24 µsec and throughput is 333.34 Gbps and 3D NoC it is 12 µsec, throughput is 666.67 Gbps, so the latency is reduced by 50% and throughput is improved by 50%. When farthest node is considered as destination node, the Memory usage and power consumption are almost same, there is 25% reduction in latency and 33.2% increase in packet Throughput. The Memory usage and power consumption are almost same, but the area to implement the design is more. Hence there is a trade-off between area and other performance parameters. Since 3-D NoCs can achieve Less Latency and High throughput, with almost same Memory usage and Power Consumption as that of 2D NoC, 3-D NoCs are said to have High QoS.