Content area
Modern Internet protocols, originally designed for modest and predictable traffic, struggle to accommodate the diverse and stringent demands of emerging applications. Despite significant advances in handling private traffic, deployable network optimizations targeting endusers remain constrained by the rigid Internet protocols and infrastructure.
This thesis investigates cross-layer network optimization opportunities for endusers by treating the Internet protocol stack as a coordinated system rather than isolated layers. Specifically, I explore how information can be propagated across application, transport, and network layers to collaboratively improve end-to-end performance without violating protocol compatibility or requiring intrusive deployment.
Guided by this cross-layer perspective, I identify three key entry points for enduser-oriented network optimization. First, I present CloudCookie, a provider-side mechanism that enables in-band information synchronization across public-facing data center traffic, facilitating flow packet scheduling and predictive load balancing. Second, I introduce a user-side prioritization-based optimization for realtime interactive streaming that leverages cross-directional dependency. Finally, I propose virtual multipath, a provider-user collaborative technique that repurposes VPN proxy nodes to emulate multipath transmission on single-interface devices by tricking the kernel’s MPTCP stack. Together, these projects demonstrate a holistic approach to enduser-oriented, cross-layer network optimization that respects Internet constraints while unlocking new performance gains through judicious use of available yet underutilized information and resources.