NoC自相似业务量生成方法研究与实现

发布时间：2018-05-20 06:42

本文选题：自相似性 + 片上网络　；参考：《西南科技大学》2017年硕士论文

【摘要】：片上网络(Network on Chip,NoC)是为解决片上系统(System on Chip,So C)通信瓶颈提出的一种全新的片上通信架构。NoC的性能高度依赖于实际业务量,而自相似性是NoC业务量的重要特性,它能直接影响NoC网络流量建模的准确程度及网络传输中的流量控制,模拟生成具有真实网络特征的业务量对NoC设计的评估和分析很关键。对比当前常用的几种自相似业务量模型,基于分形高斯过程的模型易于实现,且能更准确的描述业务量的自相似性。通过Matlab仿真对比分析两种分形高斯过程的实现方法,随机中点置换法和基于高斯噪声(Fractal Gaussian Noise,FGN)的生成方法,后者运算速度更快、准确度更高。针对已有的基于FGN的生成方法,对分形高斯噪声功率谱逼近效果差的问题,提出一种高次方线性拟合方法。将分形高斯噪声的功率谱分成有限项和无穷累加项2个部分,采用Matlab的拟合工具箱cftool对功率谱的无穷项进行高次拟合,减少运算量,去除数学推导中采用近似替代带来的误差。仿真结果表明,改进方法生成自相似业务量的相对误差比采用公式推导的方式低,生成自相似业务量的速度较快。以改进的自相似业务量生成方法为核心,设计基于FPGA的NoC自相似业务量生成器,生成器可以根据NoC节点自身的特征去设置业务量服从的参数,包括业务量的平均速率、突发性、节点注入率等。同时该模型还可以设置NoC数据包格式,包括微片长度、微片个数、源地址、目的地址等参数,支持多个工作时钟,7个通道同时输出。
[Abstract]:Network on Chip (NOC) is a new on-chip communication architecture. The performance of NOC is highly dependent on actual traffic, and self-similarity is an important characteristic of NoC traffic. It can directly affect the accuracy of NoC network traffic modeling and traffic control in network transmission. Simulation and generation of traffic with real network characteristics is critical to the evaluation and analysis of NoC design. Compared with several commonly used self-similar traffic models, the model based on fractal Gao Si process is easy to implement and can describe the self-similarity of traffic more accurately. The implementation of two fractal Gao Si processes, the random midpoint replacement method and the generating method based on Gao Si noise Fractal Gaussian Noisegnan, are compared by Matlab simulation. The latter is faster and more accurate. In order to solve the problem of poor approximation effect of fractal Gao Si noise power spectrum based on FGN, a high power linear fitting method is proposed. The power spectrum of fractal Gao Si noise is divided into two parts: finite term and infinite accumulative term. The high order fitting of infinite term of power spectrum is carried out by using Matlab fitting toolbox cftool, which can reduce the calculation amount and remove the error caused by using approximate substitution in mathematical derivation. The simulation results show that the relative error of the improved method to generate self-similar traffic is lower than that of formula derivation, and the speed of generating self-similar traffic is faster. Taking the improved self-similar traffic generation method as the core, a NoC self-similar traffic generator based on FPGA is designed. The generator can set the parameters of the traffic service according to the characteristics of the NoC node, including the average rate of traffic, which is sudden. Node injection rate and so on. At the same time, the model can also set NoC packet format, including microchip length, number of microchips, source address, destination address and other parameters, support multiple working clocks, seven channels output at the same time.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN47

【相似文献】