自相似业务量下NoC映射与缓存分配算法研究

发布时间：2018-03-22 19:40

  本文选题：片上网络　切入点：自相似业务量　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着半导体集成电路的快速发展,片上组件集成度越来越高,总线通信结构已经无法满足芯片设计的需要,为了解决这一问题,片上网络(Network on Chip,NoC)这种新型通信架构被提出。由于NoC的性能高度依赖于实际业务量,因此基于业务量开展的一系列研究具有重要意义。在NoC的自相似业务量特征下,重点研究了 NoC的映射和缓存分配问题。NoC映射结果很大程度上影响着能耗、延时等性能,当前的NoC映射研究几乎没有考虑业务量自相似特性对能耗和延时的影响。为此,在NoC自相似业务量下,从减少通信距离的角度建立能耗模型;另一方面,引入时延因子的概念建立延时模型,从系统负载分布、业务量突发强度和系统无拥塞延时三个方面对延时性能进行优化。本文结合映射问题特点使用改进的NSGAⅡ算法优化能耗和时延因子。通过几组任务流图的仿真实验表明:与传统NSGAⅡ算法、MOGA算法相比,改进算法使得能耗和延时均有明显降低,并且具有良好的收敛性。关于NoC缓存分配问题,在对路由器结构进行分析的基础上,建立一种自相似业务量下的排队延时计算模型,并基于遗传模拟退火算法设计一种缓存分配优化算法,实验表明,该算法能够极大的降低NoC输入通道的平均排队时间。
[Abstract]:With the rapid development of semiconductor integrated circuits and the increasing integration of on-chip components, the bus communication structure can no longer meet the needs of chip design. Network on Chip (NOC), a new communication architecture, is proposed. Because the performance of NoC is highly dependent on actual traffic, a series of researches based on traffic are of great significance. Under the characteristics of NoC's self-similar traffic, This paper focuses on the mapping and buffer allocation of NoC. NOC mapping results greatly affect the performance of energy consumption, delay and so on. The current research of NoC mapping hardly takes into account the effect of self-similarity of traffic on energy consumption and delay. In the case of NoC self-similar traffic, the energy consumption model is established from the angle of reducing the communication distance, on the other hand, the concept of delay factor is introduced to establish the delay model, which is based on the system load distribution. The performance of delay is optimized from three aspects: traffic burst intensity and congestion free delay. Based on the characteristics of mapping problem, the improved NSGA 鈪，

本文编号：1650095