【作者】 陶海洋；

【导师】 徐成；

【作者基本信息】 湖南大学 ， 计算机系统结构， 2009， 硕士

【摘要】 片上网络(NoC)是当前片上系统发展的一个重要方向,其中功耗和速度是设计NoC时的主要约束。随着带宽需求的增加,NoC在通信互连结构上消耗的能量所占的比例在逐步的扩大;同时,系统中处理器核的速度越来越快、数目越来越多,核之间通信的延迟将极大地影响系统整体性能。随着对高性能嵌入式计算需求的增加以及散热和电池电量有限问题的突出,能耗和延迟已经成为NoC走向实际应用必须解决的重要问题。本文针对NoC的能耗和延迟问题,通过对当前片上网络低能耗和低延迟设计技术优缺点的分析,在体系结构级通信层面上,提出了带路径标号的并行通信边求解算法来优化片上网络延迟。该算法利用NoC上各IP核之间通信的相对可预测性以及各通信事件之间的时序关系来求解各通信链路上的并发通信量,并以链路并发通信量的均衡来减少延迟。在所提算法的基础上,结合片上网络低能耗映射蚁群优化算法,本文设计了一种低能耗和低延迟综合优化方案。在蚁群优化的过程中以链路并发通信量标准偏差和能耗的乘积作为目标评价函数,实现延迟和能耗的综合优化。为了评价NoC系统的能耗和延迟综合优化效果,本文利用能耗延迟积作为衡量系统的能量-性能效率的指标。基于该指标,利用基于SystemC的NoC仿真器对文中的综合优化方案进行了仿真实验。结果表明,本文所提出的综合优化方法的能量-性能效率比链路平衡的方法优越。更多还原

【Abstract】 Network-on-chip(NoC) is an important direction of development of system on chip, and the power consumption as well as speed are the main constraints in the design of NoC. With the increase in bandwidth requirement in the NoC, the energy consumption of communication infrastructure accounts for a large proportion of the system, which is still gradually increasing; At the same time, because of the increase in terms of speed and the number of processor cores, the communication delay between these processor cores greatly affects the overall performance of the system. With the requirement for high performance embedded computing increasing, as well as the issues of heat elimination and limited battery life prominent, energy consumption and delay have became the important issues that must be addressed for NoC to practical application.This paper focuses on the energy consumption and delay of the NoC. Through the analysis of the advantages and disadvantages of the current research on low energy consumption and low delay design technology, the thesis presents an algorithm used for solving parallel communication arc with labeling in terms of the balance of concurrent traffic on communication links in the communication of the architecture level. With the relative predictability of the communication between the IP cores and the timing relationship of these communications, this algorithm solves the concurrent traffic of each communication link. Further, the balance of the concurrent traffic of the link can be used to reduce the delay. Base on the proposed algorithm and combination of low energy consumption mapping with ant colony optimization algorithm in the NoC, this thesis designs an integrated scheme for low energy consumption and low delay. Through the product of the standard deviation of the link’s concurrent traffic and energy consumption as a target evaluation function, we achieve an integrated optimization of the energy consumption and delay in the process of ant colony optimization.To comprehensively evaluate the optimization effect of the energy consumption and delay of the NoC system, this paper uses energy-delay product(EDP) as the metrics of energy-performance efficiency. Based on this metrics, we conduct some simulation experiments for the proposed integrated optimization scheme through using the SystemC-based NoC simulator. The results show that the presented method used in this paper is better than the method of link balance in terms of energy-performance efficiency.更多还原