片上网络业务量的自相似性分析及模型研究

进行片上网络的架构、映射、流控与服务质量(Quality of Service,QoS)等研究时,迫切需要一个准确的业务量模型用于延时分析与测试验证,以保证设计的性能。而现有的基于马尔科夫模型和回归模型的短程相关模型无法准确地描述业务量的突发性和分形特性,不适用于基于流水的通信信号处理片上系统(System on Chip,SoC)芯片。为了解决这个问题,通过理论与实验相结合的方法,研究了网络拓扑、任务流图、映射对业务量自相似性的影响,根据通信系统的信号处理特点建立了多处理器片上系统(Multi-core Processing System on Chip,MPSoC)数据关联模型,利用典型DSP系统进行建模实验,用实测的业务量Hurst参数拟合数据关联模型参数与Hurst参数的经验函数关系式,建立了用MPSoC数据关联模型预测和估计业务量Hurst参数的方法。实验表明,采用该业务量模型估计的Hurst参数与其真实值误差较小,能较准确地描述业务量的自相似性。

Self-similarity Analysis and Modeling for On-chip Traffic

An accurate traffic analysis model is needed for latency prediction and verification in on-chip design.Unfortunately,the state-of-art Markov-based short range dependent models cannot characterize burst and self-similarity of on-chip traffic,therefore it is not applicable for the communication and signal processing SoCs.This paper proposed a self-similar NoC traffic model based on multiple parameters to provide accurate benchmarks for the design and verification of NoC.Using theoretical derivation and experimental method,this paper established an MPSoC information relevance model,provided an empirical fitting function between the parameters of the relevance model and Hurst parameter,and established the method to estimate Hurst parameter of NoC traffic.The experimental results prove that this traffic model can achieve an approximate and effective Hurst parameter.