發(fā)布時間：2024-07-03 01:45

　　隨著互聯(lián)網(wǎng)滲透到人們生活的方方面面(物聯(lián)網(wǎng)),計算機網(wǎng)絡(luò)變得日漸龐大、復(fù)雜。在這種情況下,用不影響監(jiān)測網(wǎng)絡(luò)性能的方式獲得指標和度量值,并進行及時有效的網(wǎng)絡(luò)監(jiān)測和分析就變得至關(guān)重要。然而,測量網(wǎng)絡(luò)中所有節(jié)點的網(wǎng)絡(luò)流量是不切實際的,一種有前景的替代方案是僅在網(wǎng)絡(luò)邊緣進行測量,并從這些測量值中推斷網(wǎng)絡(luò)的內(nèi)部行為。為了解決內(nèi)部鏈路參數(shù)測量(例如時延和丟包率)的問題,本文采用網(wǎng)絡(luò)層析(NT)技術(shù),收集基于端到端測量的路徑性能數(shù)據(jù),然后使用統(tǒng)計計算的方法推斷邏輯鏈路性能的概率分布。這種從端到端估計鏈路性能的技術(shù)既不需要內(nèi)部網(wǎng)絡(luò)的協(xié)作,也不依賴通信協(xié)議。此外,本文在網(wǎng)絡(luò)層析成像技術(shù)上融合了一種新的統(tǒng)計方法,使得建模網(wǎng)絡(luò)更容易估計內(nèi)部鏈路性能參數(shù)的性能。這種方法就是機器學(xué)習(xí)(ML),尤其是線性回歸模型。該技術(shù)能夠在給定輸入值(例如路徑時延)后預(yù)測真實值(例如鏈路時延),比如說讓模型從給定的樣本數(shù)據(jù)(學(xué)習(xí)數(shù)據(jù)大約占總數(shù)據(jù)的80%)中學(xué)習(xí),然后使用20%的數(shù)據(jù)(測試數(shù)據(jù))驗證模型。將估計得到的時延值與使用網(wǎng)絡(luò)模擬器NS2對一個有線網(wǎng)絡(luò)仿真生成的實際時延值進行比較,以兩者的計算誤差值(均方誤差MSE)評估模...

【文章頁數(shù)】：61 頁

【學(xué)位級別】：碩士

【文章目錄】：
摘要
ABSTRACT
LIST OF ABBREVIATIONS
Chapter 1 Introduction
    1.1 Introduction
        1.1.1 Network Tomography in the Internet
    1.2 Problem Statement
    1.3 Motivation
    1.4 Contribution
    1.5 Thesis Layout
Chapter 2 Literature Review
    2.1 Introduction
        2.1.1 Computer Network
        2.1.2 Tomography
        2.1.3 Network Tomograph
    2.2 Delay Network Tomography
        2.2.1 Estimation of Propagation Link Delay
        2.2.2 Estimation of Link Delay Density
    2.3 Traffic Network Tomography
        2.3.1 Vardi's model
        2.3.2 Vanderbei and lannone's model
        2.3.3 Maximum likelihood parameter estimation
        2.3.4 The EM algorithm
        2.3.5 Other aspects for the Network Tomography
Chapter 3 Statistica Inference
    3.1 Introduction
    3.2 Categories of Machine Learning
        3.2.1 Supervised Learning
        3.2.2 Unsupervised Learning
        3.2.3 Reinforcement Learning
    3.3 Linear Regression
        3.3.1 Simple Regression
        3.3.2 Multiple Regression
    3.4 The Loss Function
Chapter 4 Methodology
    4.1 Introduction
    4.2 Simulation methodology
        4.2.1 Capabilities and limitations of ns2
        4.2.2 Network simulation using ns2
        4.2.3 Network Animator NAM
    4.3 The Designed Model
        4.3.1 Machine Learning implementation
    4.4 End-To-end Delay
        4.4.1 The measurement procedure
Chapter 5 Evaluation and Results
    5.1 Introduction
    5.2 Result and Discussion
Chapter 6 Conclusion and Future Work
    6.1 Conclusion
    6.2 Future Work
Bibliography
ACKNOWLEDGEMENT
PUBLICATIONS AND MASTER ACTIVITIES



本文編號：4000335