论文部分内容阅读
抽样方法广泛地应用于网络测量与其他领域对被测总体的指标进行估计.研究表明,多种网络指标呈现重尾分布或自相似的特征.这些特性为准确估计总体指标带来了诸多困难.但同时,对被测网络指标进行建模也有着重要的应用.然而,建立精确网络模型是困难的.从时间序列拟合角度出发,提出了一种基于拟合的自适应抽样方法,对被测指标进行基于测量的建模.工作主要体现在: (1) 采用分段线性函数对被测指标进行逼近,建立基于测量的模型; (2) 与常用的抽样方法相比,在相同的样本数情况下,由拟合模型对指标进行的估计更准确、更稳定;通过对两个测量记录的分析表明,在与常用抽样方法保持相同的拟合误差时,自适应抽样方法明显地减少了所需采集的样本数量; (3) 与其他概率抽样方法相比,自适应抽样最终抽取的样本数更稳定、更可靠,并给出了最终样本数的概率分布.
Sampling methods are widely used in network surveying and other fields to estimate the overall measured indicators.Research shows that a variety of network indicators show the characteristics of heavy-tailed distribution or self-similar These characteristics for the accurate estimation of the overall index brings many difficulties. But at the same time, it is also important to model the measured network index.However, it is difficult to establish an accurate network model.From the perspective of time series fitting, a fitting-based adaptive sampling method is proposed, Measure index based on the measurement of modeling.Main work is reflected in: (1) using piecewise linear function of the measured index approximation, the establishment of measurement-based model; (2) compared with the commonly used sampling method, the same sample The number of cases, the fitting model to measure the indicators more accurate and more stable; through the analysis of two measurement records show that in the sampling method with the same to maintain the same fitting error, adaptive sampling method significantly reduced The number of samples needed to be collected; (3) Compared with other probability sampling methods, the number of samples finally extracted by adaptive sampling is more stable and reliable, and the final number of samples Rate distribution.