Research On Some Key Issues For High-Speed Internet Performance Measurements

The rapid development of network technologies, especially the wide application of TCP/IP protocol-based Internet deeply reforms our working, learning and living actives. Internet acting as an information infrastructure has been becoming the infrastructure of the whole society. Due to the inherited high isomerizationess, complexitiness and openness of Internet, the evolution of network technologies and network-based applications are also faced with some acute problems. Network control and management, services guarantee and thoroughly understanding network characteristics have become even more difficult.Research on network measurement methodologies is an important filed for network related researches in last decade, and has been a hot spot in recent years. The aim of network measurement is to evaluate the operation of network, develop measurement-based analytical model for network metrics, lift the performance of underlying communication networks and optimize the network system design and flow engineering. In the thesis, we survey the state of the art in network performance measurements, including motivations, metrics architecture and common performance metrics in the context of Internet environment. We review the past progresses and important achievements in this research area. The classification and characters of measurement methodologies are summarized. The criteria for valid measurement method designing and evaluation are described in detail. We also point out some key issues for large scale, high-speed network performance measurements. This thesis presents some key issues of Internet performance measurements from large scale, high-speed network point of view and results in following innovative achievements:Proposing an efficient algorithm for end system clocks synchronizationBased on the analysis of end system clock activities and its abnormalities, we systematically propose the relative clock dynamic model between end-to-end system clocks for the first time. By constructing the one-way delay time series via periodically probed packets between end systems, and the relative clock dynamic model, we develop a new time series segmentation criteria and an optimized segmentation algorithm to accomplish automatic time series clustering. After the one-way delay time series is clustered, it is easy to detect the clock dynamics in the measurement duration, remove the clock skew between end systems and achieve relative clock synchronization. A sliding window based online clock synchronization algorithm is also developed. The clock synchronization algorithm removes those acute assumptions on system clock activities in related works and is of more robust and practicable. We have held a patent on this achievement (Application No.: 03127752.7, Publication No.: 1484410, Publication Data: 2004.3.24).Proposing a Fitting-based Adaptive Sampling MethodologyFrom the accurate time series approximation point of view, we propose a Fitting-based Adaptive Sampling Methodology (FASM). The FASM algorithm adaptively adjusts the time interval between consecutive samples according to the evolution trend of interested metric which is estimated from the obtained samples. By constructing the piecewise linear function from samples, it is able to approximate the measured metric more precisely. Compared with current popular sampling algorithms, FASM algorithm can provide more precise estimation of metric under the same samples. In the condition of retaining the same approximation errors, our adaptive sampling methodology only requires fewer samples, which will save sampling costs considerably during sampling process. At the same time, the number of practically obtained samples by FASM algorithm is more stable and reliable compared against other probability based sampling methodologies. The probability distribution of obtained sample size is also derived theoretically. We have held a patent on this achievement (Application No.: 03133100.9, Publication No.: 1477554, Publication Data: 2004.2.25).Proposing a Probability-Guaranteed Ada...