Graduation Semester and Year
Fall 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science
Department
Computer Science and Engineering
First Advisor
Hao Che
Second Advisor
Hong Jiang
Third Advisor
Song Jiang
Fourth Advisor
Yonghe Liu
Abstract
Multipath TCP (MPTCP), as a standards-based Internet transport protocol, seeks to leverage path diversity to enhance throughput and reliability. Yet, despite a decade of development and deployment, existing MPTCP designs remain fundamentally cost-unaware and often yield only marginal throughput improvements over TCP. This dissertation establishes rigorous theoretical foundations and presents deployable, end-to-end algorithms that directly address these long-standing limitations.
To resolve the cost-unaware nature of current MPTCP deployments, we introduce \textbf{Uni-MPTCP}$(\boldsymbol{\omega}, n)$, a unified and principled cost-aware congestion-control framework derived from the Network Utility Maximization (NUM) formulation. By integrating a tunable weight vector $\boldsymbol{\omega}$, Uni-MPTCP enables systematically controlled cost-sensitive rate allocation while provably preserving fairness, resource pooling, and other desired MPTCP properties. A two-path instantiation is shown to achieve performance that is nearly indistinguishable from the NUM-optimal benchmark, thereby validating the analytical foundation.
To overcome the limited throughput gains of existing MPTCP algorithms, we first establish that NUM-optimal solutions can theoretically exceed the throughput of current MPTCP by at least \textbf{17\%}. Motivated by this fundamental performance gap, we design \textbf{N-MPTCP}$(\boldsymbol{\omega})$, a lightweight, measurement-driven congestion-control algorithm that closely approximates the NUM-optimal operating point. N-MPTCP not only realizes the theoretically predicted throughput improvement in practice, but does so with a deployable end-to-end design that requires no explicit network support.
Comprehensive ns-3 simulation studies demonstrate that Uni-MPTCP and N-MPTCP substantially extend the achievable performance region of MPTCP across a wide range of network conditions. Collectively, the theoretical insights and practical designs developed in this dissertation advance the state of the art in multipath transport by providing the first cost-aware and near-NUM-optimal MPTCP solutions that are both analytically grounded and deployable in real-world environments.
Keywords
Multipath Congestion Control, Network Utility Maximization, Convex Optimization, KKT Conditions, TCP Fairness Theory, Throughput Optimality, Coupled Congestion Control, Resource Pooling Theory, AIMD Dynamics, Heterogeneous Path Modeling
Disciplines
OS and Networks | Theory and Algorithms
License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Wang, Xuan, "MPTCP Design Principles Revisited: Network Utility Maximization-based Multipath Traffic Control" (2025). Computer Science and Engineering Dissertations. 426.
https://mavmatrix.uta.edu/cse_dissertations/426