Enhanced reliable and adaptive Routing models for networks on chip

Abstract

The main objective of this research is to develop a reliable and newlineCongestion aware adaptive routing model to improve the performance of the newlineNetwork on Chip NoC Due to limited buffering resources in the wormhole newlineswitching technique the congestion in the downstream routers propagates to newlineupstream routers which severely cause the entire network to be congested newlineThe impact of buffers on the performance of Mesh architecture for both newlineunicast and multicast routing algorithms have been dealt with An efficient newlineGreedy algorithm is proposed for unicast routing which automatically newlineallocates the buffering resources to different input NoC channels such that newlinethe communication performance is improved while satisfying the total newlinebuffering resource capacities newlineThe developed dynamic buffer allocation algorithm customizes the newlinebuffer required within the given buffer capacity based on the application newlinerequirement In order to identify the possible channels of the node for a newlineparticular application that will get congested a queuing model is proposed newlineThe traffic arrival process at each node is assumed to be Poisson distribution newlineEach input channel buffer in the network is modeled as an MM1K finite newlinequeue to find the probability of the channel that will become full By varying newlinethe buffer size according to the traffic pattern the proposed algorithm is able newlineto generate systems with much better performance while using 20 percent less newlinebuffering resources compared to uniform buffer allocation newline