PhD Proposal: End-to-end measurement and analysis of real-time communication systems

Real-time communication (RTC) has become fundamental to modern digital interaction, yet commercial systems remain largely opaque in how they implement protocols, deploy relay infrastructures, and support emerging AI-driven conversational experiences. This thesis investigates RTC systems from three complementary perspectives to advance both empirical understanding and future design. First, we perform the first cross-application protocol compliance study of popular RTC services, revealing widespread deviations from STUN, TURN, RTP, and RTCP specifications and highlighting challenges for interoperability and protocol evolution. Second, we conduct the first global-scale measurement of commercial relay infrastructures, mapping relay deployments for Zoom, WhatsApp, Messenger, and Discord across globally distributed vantage points. We uncover radically different relay-placement strategies, relay-targeting policies, and fallback behaviors, and quantify their impact on end-to-end latency. Finally, we look ahead to next-generation Gen-AI calling, where large language models interact with users through continuous audio streams. We introduce personalized semantic endpointing, combining user-conditioned models, prosody-aware representations, and reinforcement learning to improve responsiveness while reducing conversational interruptions. Together, these works build a comprehensive, systems-driven understanding of today’s RTC ecosystem and lay the technical foundation for more interoperable, performant, and human-centered real-time communication systems.