by Chung Ling Chi. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2001. / Includes bibliographical references (leaves 74-76). / Abstracts in English and Chinese. / Chapter Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Overview --- p.1 / Chapter 1.2 --- Motivation and Proposed Solution --- p.2 / Chapter 1.3 --- Organization of the Thesis --- p.4 / Chapter Chapter 2 --- Background --- p.5 / Chapter 2.1 --- Basics of Transmission Control Protocol --- p.5 / Chapter 2.1.1 --- Slow Start and Congestion Avoidance --- p.5 / Chapter 2.1.2 --- Fast Retransmit and Fast Recovery --- p.7 / Chapter 2.2 --- SACK TCP Mechanism --- p.8 / Chapter 2.2.1 --- SACK-permitted Option during Three-way Handshake --- p.8 / Chapter 2.2.2 --- SACK blocks in SACK Option --- p.9 / Chapter 2.2.3 --- Interpreting the SACK Option using Scoreboard --- p.10 / Chapter 2.2.4 --- Retransmission Strategy --- p.11 / Chapter 2.3 --- TCP Veno Mechanism --- p.13 / Chapter 2.3.1 --- Refined Additive Increase --- p.13 / Chapter 2.3.2 --- Refined Multiplicative Decrease --- p.14 / Chapter Chapter 3 --- SACK TCPVeno --- p.16 / Chapter 3.1 --- Distinguishing between Types of Packet Loss --- p.17 / Chapter 3.2 --- Refined Multiplicative Decrease --- p.21 / Chapter 3.2.1 --- Algorithm --- p.21 / Chapter 3.2.2 --- Recovery in Consecutive packet Losses --- p.22 / Chapter 3.2.3 --- Recovering Multiple Packet Losses within a Single Window --- p.26 / Chapter 3.3 --- Refined Additive Increase --- p.37 / Chapter 3.3.1 --- Algorithm --- p.37 / Chapter 3.3.2 --- Advantages --- p.40 / Chapter 3.4 --- Other Issues --- p.43 / Chapter 3.4.1 --- Two Side Modifications --- p.43 / Chapter Chapter 4 --- Experiments --- p.44 / Chapter 4.1 --- The Network Scenario --- p.44 / Chapter 4.1.1 --- Dummynet --- p.45 / Chapter 4.2 --- Experiment Results --- p.47 / Chapter 4.2.1 --- Single Connection --- p.47 / Chapter 4.2.1.1 --- Congestion Window Evolution --- p.47 / Chapter 4.2.1.2 --- Sending Rate and Throughput Evolution --- p.49 / Chapter 4.2.1.2.1 --- Impact of Packet Loss Rate Due to Lossy Link --- p.49 / Chapter 4.2.1.2.2 --- Impact of Buffering --- p.52 / Chapter 4.2.1.2.3 --- Impact of Propagation Delay --- p.57 / Chapter 4.2.2 --- Multiple Connections --- p.62 / Chapter 4.2.2.1 --- Fairness --- p.62 / Chapter 4.2.2.2 --- Compatibility --- p.67 / Chapter Chapter 5 --- Conclusion --- p.72 / Bibliography --- p.74

近年來，使用者通過移動數據網路，如3G和LTE，連接到互聯網的數目急劇增加。眾所周知無線網路和移動數據網路展現的特點和有線網路有很大的不同。儘管如此，大多數移動應用程式的基本構建塊，即傳輸控制協議（TCP），在很大程度上仍是根植於有線網路。本論文通過廣泛的開展多個移動數據網路，包括3G，HSPA，最新的LTE網路的測試和實驗，探討TCP在現代移動數據網路的性能。儘管移動數據網路頻寬的迅速增加，我們的測量結果均顯示，現有的TCP實現在實踐中表現不佳，未能利用高速移動數據網路豐富的頻寬。這項工作解決TCP的性能限制，採用一種新的方法透明協議優化，通過在中間網路設備即時優化TCP，顯著提高TCP的吞吐量。具體來說，這項工作發展（一）一個新穎的機會傳輸算法克服TCP的流量控制的瓶頸;（二）一個傳輸速率控制演算法來解決TCP的拥塞控制的瓶頸;（三）一個新穎的投機重傳演算法，以提高TCP在重傳中的吞吐量;（四）用隨機模型來量化TCP吞吐量性能對移動網路資源利用率的影響;（五）一個新的隊列長度測量算法，為擁塞控制和網路監測打開一條新的途徑。另外，擬議的協議優化技術已全面實施，變成一個移動加速器裝置已經成功在三個不同的3G/LTE生產移動數據網路領域試用，實驗顯示TCP的吞吐量從48%增加至163%。在發明一種新的傳輸協議，或修改現有的TCP實施相比，所提出的方法不要求在用戶端/伺服器的主機現有的TCP實施任何修改，不需要重新配置伺服器或用戶端，並因此可以容易在現今的3G和4G移動網路部署，提高所有現有網路上運行在TCP之上的應用程式的吞吐量性能。 / The number of Internet users which are connected via mobile networks such as 3G and LTE has increased dramatically in recent years. It is well-known that wireless networks in general, and mobile data networks in particular, exhibit characteristics that are very different from their wired counterparts. Nevertheless, the fundamental building block of most Internet applications, namely the Transmission Control Protocol (TCP), is still largely rooted in wired networks. This dissertation investigate the performance of TCP over modern mobile data networks through extensive measurements and experiments carried out in multiple production data networks, ranging from 3G, HSPA, to the latest LTE networks. Despite the rapid increases in mobile network bandwidth, our measurements consistently reveal that existing TCP implementations perform sub-optimally in practice, failing to utilize the abundant bandwidth available in high-speed mobile networks. This work tackles the performance limitations of TCP using a novel approach - transparent protocol optimization, to significantly improve TCP’s throughput performance using on-the-fly protocol optimization carried out by an intermediate network device in-between the TCP end-hosts. Specifically, this work develops (i) a novel opportunistic transmission algorithm to overcome the TCP’s flow control bottleneck; (ii) a transmission rate control algorithm to tackle TCP’s congestion control bottleneck; (iii) a new opportunistic retransmission algorithm to improve TCP’s performance during packet loss recovery; (iv) a stochastic model to quantify the impact of TCP throughput performance on mobile network capacity; and (v) a new queue length estimation algorithm which opens a new avenue for congestion control and network monitoring. In addition, the proposed protocol optimization techniques have been fully implemented into a mobile accelerator device which has been successfully field trialed in three different production 3G/LTE mobile networks, consistently increasing TCP’s throughput by 48% to 163%. In contrast to inventing a new transport protocol or modifying an existing TCP implementation, the proposed approach does not require any modification to the existing TCP implementation at the client/server hosts, does not require any reconfiguration of the server or client, and hence can be deployed readily in today’s 3G and 4G mobile networks, raising the throughput performance of all existing network applications running atop TCP. / Detailed summary in vernacular field only. / Liu, Ke. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2013. / Includes bibliographical references (leaves 166-174). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts also in Chinese. / Abstract --- p.2 / Acknowledgement --- p.6 / Chapter 1 --- p.1 / Introduction --- p.1 / Chapter 1.1 --- Motivation --- p.1 / Chapter 1.2 --- Contributions --- p.3 / Chapter 1.3 --- Structure of the Thesis --- p.6 / Chapter 2 --- p.9 / Flow and Congestion Control --- p.9 / Chapter 2.1 --- TCP Performance Bottlenecks --- p.9 / Chapter 2.2 --- Background and related works --- p.16 / Chapter 2.3 --- Transparent Protocol Optimization --- p.20 / Chapter 2.3.1 --- Opportunistic Transmission --- p.20 / Chapter 2.3.2 --- Transmission Rate Control --- p.22 / Chapter 2.3.3 --- Lost Packet Recovery --- p.27 / Chapter 2.4 --- Modeling and Analysis --- p.28 / Chapter 2.4.1 --- Background and Assumptions --- p.28 / Chapter 2.4.2 --- Queue Length at the Radio Interface --- p.31 / Chapter 2.4.3 --- Queue Length Bounds --- p.38 / Chapter 2.4.4 --- Guaranteeing Full Bandwidth Utilization --- p.45 / Chapter 2.4.5 --- Link Buffer Size Requirement --- p.47 / Chapter 2.5 --- Performance Evaluation --- p.53 / Chapter 2.5.1 --- Parameter Tuning --- p.53 / Chapter 2.5.2 --- Bandwidth Efficiency --- p.56 / Chapter 3 --- p.62 / Packet Loss Recovery --- p.62 / Chapter 3.1 --- Introduction --- p.62 / Chapter 3.2 --- TCP Loss Recovery Revisited --- p.64 / Chapter 3.2.1 --- Standard TCP Loss Recovery Algorithm --- p.64 / Chapter 3.2.2 --- Loss Recovery Algorithm in Linux --- p.66 / Chapter 3.2.3 --- Loss Recovery Algorithm in A-TCP --- p.67 / Chapter 3.3 --- Efficiency of TCP Loss Recovery Algorithms --- p.68 / Chapter 3.3.1 --- Standard TCP Loss Recovery Algorithm --- p.70 / Chapter 3.3.2 --- TCP Loss Recovery in Linux --- p.72 / Chapter 3.3.3 --- Loss Recovery Algorithm Used in A-TCP --- p.72 / Chapter 3.3.4 --- Discussions --- p.73 / Chapter 3.4 --- Opportunistic Retransmission --- p.74 / Chapter 3.4.1 --- Applications and Performance Analysis --- p.76 / Chapter 3.4.2 --- Bandwidth Utilization During Loss Recovery --- p.78 / Chapter 3.5 --- Experimental Results --- p.81 / Chapter 3.5.1 --- Model Validation --- p.85 / Chapter 3.5.2 --- Impact of Loss Recovery Phase on TCP Throughput --- p.85 / Chapter 3.5.3 --- A-TCP with Opportunistic Retransmission --- p.86 / Chapter 3.6 --- Summary --- p.87 / Chapter 4 --- p.89 / Impact on Mobile Network Capacity --- p.89 / Chapter 4.1 --- Introduction --- p.89 / Chapter 4.2 --- Background and Related Work --- p.91 / Chapter 4.2.1 --- TCP Performance over Mobile Data Networks --- p.91 / Chapter 4.2.2 --- Modeling of Mobile Data Networks --- p.92 / Chapter 4.3 --- System Model --- p.94 / Chapter 4.3.1 --- Mobile Cell Bandwidth Allocation --- p.95 / Chapter 4.3.2 --- Markov Chain Model --- p.96 / Chapter 4.3.3 --- Performance Metric for Mobile Internet --- p.98 / Chapter 4.3.4 --- Protocol-limited Capacity Loss --- p.100 / Chapter 4.3.5 --- Channel-limited Capacity Loss --- p.101 / Chapter 4.4 --- Performance Evaluation --- p.102 / Chapter 4.4.1 --- Service Response Time --- p.103 / Chapter 4.4.2 --- Network Capacity Loss --- p.105 / Chapter 5 --- p.114 / Mobile Link Queue Length Estimation --- p.114 / Chapter 5.1 --- Introduction --- p.115 / Chapter 5.2 --- Sum-of-Delay (SoD) algorithm Revisited --- p.117 / Chapter 5.2.1 --- Queue Length and Link Buffer Size Estimation --- p.117 / Chapter 5.2.2 --- A Bound on Estimation Error --- p.120 / Chapter 5.2.3 --- Impact of Uplink Delay Variations --- p.122 / Chapter 5.3 --- Uplink Delay Variation Compensation --- p.127 / Chapter 5.3.1 --- Exploiting the TCP Timestamp Option --- p.127 / Chapter 5.3.2 --- TCP Timestamp Granularity --- p.130 / Chapter 5.4 --- Performance Evaluation --- p.131 / Chapter 5.4.1 --- Link buffer size estimation under uplink delay variations --- p.132 / Chapter 5.4.2 --- Queue length estimation under uplink delay variations --- p.136 / Chapter 5.5 --- Summary --- p.136 / Chapter 6 --- p.139 / Summary and Future Works --- p.139 / Chapter 6.1 --- Transparent Protocol Optimization --- p.139 / Chapter 6.2 --- Cross-Layer Modeling and Optimization of Mobile Networks --- p.141 / Chapter Appendix A. --- Derivation of Equations (2.24) and (2.25) --- p.143 / Chapter Appendix B. --- Proof of Theorem 2.1 --- p.145 / Chapter Appendix C. --- for Proof of Theorem 2.2 --- p.147 / Chapter Appendix D. --- for Proof of Theorem 2.3 --- p.150 / Chapter Appendix E. --- for Proof of Theorem 2.4 --- p.151 / Chapter Appendix F. --- for Proof of Theorem 2.5 --- p.152 / Chapter Appendix G. --- for Proof of Theorem 2.6 --- p.153 / Chapter Appendix H. --- for Proof of Theorem 2.7 --- p.156 / Chapter Appendix I. --- for Proof of Theorem 2.8 --- p.157 / Chapter Appendix J. --- for Proof of Theorem 3.2 --- p.161 / Chapter Appendix K. --- for Theorem 3.4 --- p.163 / Chapter Appendix H. --- for Theorem 3.5 --- p.164 / Bibliography --- p.166

by Fu Chengpeng. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2001. / Includes bibliographical references (p. 102-119). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Mode of access: World Wide Web. / Abstracts in English and Chinese.

In 3GPP Release 5 and beyond, an All-IP architecture has been specified. This indicates that convergence of mobile applications such as voice, video and gaming to IP is not a "trend" anymore, but a reality. IP mobility has therefore been intensively studied in recent years. Majority of the existing IP mobility schemes, including Mobile IPv6 (MIPv6), the current de facto standard solution for IP mobility, are designed around a two-tier addressing model. In this model, while visiting a foreign link, a mobile node (MN) is identified by its home address assigned by its home link but is located by the care-of-address (CoA) acquired from the foreign link. Incoming packets for the MN are routed to its home link as usual, but are intercepted by the home agent and tunneled to the CoA. This model is simple and is well accepted. However, when it comes to real-time applications, it also has been known to be ineffective in terms of handoff delay and bandwidth consumption due to, respectively, its lengthy CoA acquisition and the extra IP header for tunneling. The latter is especially expensive for the case of real-time applications because of the excessive overhead induced by the extra IP header (20 bytes for IPv4 and 40 bytes for IPv6) to the packet payload size (∼20-160 bytes). / In this thesis, we show that (i) can be overcome when a direct Layer-3 connection between the home and any particular visiting domain is available so that inter-domain routing effectively becomes routing within the same logical hierarchy. We call a global network formed by the directly Layer-3 connected domains the Global Universal Addressing (GUA) framework. When deployed on the GUA framework, the existing local mobility schemes can easily be upgraded to support global mobility as seamlessly as local mobility with no modification needed. / Much work has been devoted to improving the two-tier addressing model, including various local mobility schemes such as HAWAII and Cellular IP. These schemes eliminate the CoA acquisition when the MNs move within one domain, but revert back to the two-tier addressing model when the mobility is across different domains (or so-called global mobility). These schemes therefore inherit all the drawbacks of the two-tier addressing model when it comes to global mobility. It has been argued that mobility across domains is rare. However, looking into the near future, this assumption is certainly not applicable to the upcoming fourth-generation (4G) wireless architecture in which the MNs can dynamically choose the best connected wireless interface among heterogeneous networks (e.g., WiFi, WiMax, etc.) of different domains as they move. Therefore, an efficient solution is needed to handle the frequent inter-domain mobility, or global mobility, in the form of heterogeneous handoffs as well. / To address (ii), we propose a new IP mobility scheme called Cellular Universal IP (CUIP), which runs on the GUA framework and makes use of a home route concept also proposed in this thesis. The home route concept intelligently integrates the efficiency of prefix routing and flexibility of full-address routing to achieve high performance and routing scalability under the universal addressing model. In addition, based on IPv6, CUIP makes use of the IPv6 option header to embed the route-update information of an MN in the outgoing data packets for a short period after handoff, so that global routing information can be effectively updated along the path traversed by the packets. We study the performance of CUIP quantitatively and show the following: (1) the average number of routers updated per handoff is less than three, so that the average handoff delay is minimal. (2) The routing table complexity is asymptotically independent of the depth and monotonically decreasing with the width of the network hierarchy. That is, routing scalability is not a concern even in large networks. / To efficiently support global mobility, a universal addressing model, under which a mobile node is always identified and located by the same IP address globally, is an obvious answer to the problems associated with the two-tier addressing model. However, the universal addressing model has been considered to be infeasible due to difficulties in (i) inter-domain (or cross-prefix) IP routing and (ii) routing table scaling. / by Lam, Pak Kit. / "June 2007." / Adviser: Soung Liew. / Source: Dissertation Abstracts International, Volume: 69-01, Section: B, page: 0553. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2007. / Includes bibliographical references (p. 128-130). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese. / School code: 1307.

Thesis (M.S.)--Worcester Polytechnic Institute. / Keywords: networks; performance evaluation; streaming media; measurement. Includes bibliographical references (p. 89-93).

Thesis (M.S.)--West Virginia University, 2004. / Title from document title page. Document formatted into pages; contains ix, 109 p. : ill. (some col.). Includes abstract. Includes bibliographical references (p. 56-58).

Thesis (M.S.)--Worcester Polytechnic Institute. / Keywords: AECN; heterogeneous flows; RED; ECN; goodput; fairness. Includes bibliographical references.

Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (leaves 99-106).

Thesis (M.S.)--University of Florida, 2002. / Title from title page of source document. Includes vita. Includes bibliographical references.

Thesis (M.S.)--University of Florida, 2002. / Title from title page of source document. Includes vita. Includes bibliographical references.