Research On Traffic Modeling Of Network On Chip And Applications

When the more and more IP modules are integrated into a single chip, the bus based on chip communication schemes become the bottleneck for the performance of the Chips. Thus, network on chip(No C) architectures have been proposed to replace global interconnects. On the other hand, the performance of No C is highly dependent on the practical traffic pattern. So the traffic models is one of the issues for the No C design, including the network topologies, protocols and performance evaluation implementations. In this dissertation, the traffic model and application are studied for No C platform, including the method of traffic estimation and generation based on the self-similarity, the approach of multi fractal traffic estimation and generation, global traffic modeling based on topology in No C and the buffer allocation optimization algorithm of No C. The main contributions of this dissertation are listed as follows:In chapter 2, we study five Hurst index estimation algorithms based on R/S, aggregated variance, periodgram, Whittle and Wavelet, respectively. We also perform the accuracy testing for these five methods. Then, we explain the causes of traffic self-similarity in No C. Lastly, we propose two self-similar No C traffic generation methods, the improved self-similar traffic generator based on RMD(ISTG-RMD) and the improved arbitrary marginal distribution self-similar traffic generator based on Fourier transform( IAMDSTG-FT). The simulation results prove that theses two traffic generation methods can achieve an approximate and effective No C data traffic.In chapter 3, we design the method of multi fractal spectrum estimation with multi wavelet according to the characters of multi fractal process. We propose an approach to generate accurately No C multi fractal traffic based on multi fractal wavelet model(Mf WM). We also analyze the advantages of multi fractal theory for No C traffic compared with self-Similar traffic model. In the end of the chapter, a hardware No C traffic generator is provided which can generates any No C traffic for testings and simulations.In chapter 4, we use self-similar theory and multi fractal theory to get five conclusions for No C traffic, which can help designers to analyze the traffic characters for the specific applications. According to the experimental statistical methods, this we give a systolic unit polynomial fitting traffic model(SUPFTM) which provides a distinct function relationship of in and out traffic. Simulation results show that the proposed SUPFTM method is accurate and effective. In the last, a multi task No C global traffic model for the process nodes is proposed which can provied No C designers an approach for global traffic simulation in No C testing and evaluating.In chapter 5, we firstly analyze the routers buffer resources optimization for the No C design according to the traffic model. We propose a novel method of buffer allocation based on self-similar No C traffic. The case study proves that the proposed buffer allocated method achieves lower queuing delay than traditional uniformly buffer allocated. At the end of this chapter, the critical path buffer allocated method is also studied deeply. Simulation results shows that the critical path with proposed buffer allocated method can be very effectively reduced due to the reeducation of queuing delay for the data packets.In the end, we provide a demo for the algorithms which are proposed in this dissertation based on a complexity adaptive communication receiver with No C architecture. We decompose the receiver into sixteen sub tasks and evaluate the commutation data rate and Hurst parameters for each sub tasks. Then, we use the buffer allocated method proposed in chapter 5 to get the best buffer resources configuration plan in the implementation of adaptive communication receiver.