Research On Loss Inference Algorithm Based On Network Tomography

As the kinds and amount of Internet traffic have expanded rapidly in these years, the need for the QoS on the network is ever increasing. How to get the optimal performance on the existing infrastructure is a problem cared heavily by network manager and ISP. In order to save the above problems, people need to understand the status and performance distribution of network for designing and managing network more reasonable or highly effective. Network tomography is an emerging network research area in recent years. They use end-to-end performance measurement result to derive network internal performance distribution or topology, and have pointed out a new direction for the network performance and topology analysis.In this paper, at first we introduce the purpose, significance, domestic and foreign research present situation of network tomography. We give a detailed presentation on tomography and measure methods by different classified method.After introduced the elementary knowledge about tomography, we elaborate in detail the derivative process of multicast-based Direct Estimator Algorithm, Expectation Maximum Algorithm and unicast-based loss inference Method. After that, we describe the simulated environment in which we research tomography technology. According network traffic model, we construct the background and detection traffic, and analyze each characteristic. Simultaneously used the massive experimental result, we determine important parameters of detection traffic as packets sending gap and packet type.What flow is that, we use model simulation to analyze the EM algorithm contraction ratio. Through network simulating, we analyze the factors which can influence loss inference algorithm accuracy like measurement strategy or routing algorithm. We analyze the accuracy and contraction characteristic of multicast-based Direct Algorithm and EM Algorithm, and compare the error factor between them. Obtain the conclusion that EM Algorithm is better than Direct Algorithm. Aimed at measuring the large-scale network, we analyze multicast-based loss inference used multiple points and introduce minimum variance weighted average and EM algorithms, and compare the difference on accuracy and...