Study Of NoC Backlog Bound Optimization Based On Network Calculus

With the development of integrated circuits,NoC performance analysis has become one of the hotspots in NoC research;the worst case analysis of NoC performance is of the most important one.Existing studies mainly focus on analyzing the delay bound,rarely on the upper bound rare,which is important for evaluating the performance and costs of NoC.In this thesis,based on the network calculus,combined with the firefly algorithm,we calculate and optimize the upper bound of the network.The main contribution is as follows:(1)Deriving the backlog bound calculation model of NoCUsing network calculus,an analytical model for computing both the global network backlog and the backlog of single node backlog on NoC has been derived.The backlog of a single stream is calculated and then assigned to each node after the selection.The backlog matrix is established using the backlog bound of each node which is the highest backlog of all flow going through current node.The global backlog bound of the network is calculated,through the backlog matrix.(2)Deriving the optimization algorithm for single node backlog boundFor the problem of congestion and delay of network caused by the excessive backlog bound of a single node backlog,the backlog bound of single node is obtained by the method in(1).Taking single node backlog bound as objective function,we optimize the single node bound through firefly algorithm in different network performance parameters with applications such as VOPD,MWD 263 ENC,MP3DEC VOPD,MP3 ENC MP3DEC,and 263 DEC MP3DEC.The experimental results show that,compared with the prior optimization,the optimization of the node backlog bound is 30.26%,51.04%,and 30.31%,respectively.(3)Deriving the optimization algorithm for global backlog boundFor the problem of area redundancy and resource consumption caused by global backlog bound,the global backlog bound is obtained by the method in(1).Taking global backlog bound as objective function,we optimize the global backlog bound through firefly algorithm in different network performance parameters with VOPD application.The experimental results show that,compared with the prior optimization,the optimization of the global backlog bound is 2-4%.