Dynamic Media Streaming under Predictable Reliability

Abstract:

As audiovisual communications services become the dominant traffic source in the Future Internet, their stable coexistence and their fairness towards background traffic is an active research topic. The Internet implements flow multiplexing in a decentralized fashion via congestion control. Audiovisual media streams, however, are classified as inelastic traffic flows that do not allow rate control. Today's Internet streaming applications implement dynamic HTTP streaming in order to adapt to the transport protocol's congestion window size. Nevertheless, the infeasibility of TCP's congestion and error control for low-latency, high-rate communications services leads to significant underutilization of the available bandwidth. As a result, the multimedia quality suffers unnecessarily. In order to address TCP's drawbacks, we implement a novel transport protocol - Predictably Reliable Real-time Transport (PRRT), a protocol layer that efficiently supports the reliability required by multimedia services under their specific time constraint. PRRT is based on hybrid error coding that allows for both proactive and reactive reliability mechanisms. Specifically, it is designed to achieve the optimal trade-off between those mechanisms under strict delay constraints in order to optimize bandwidth utilization. Based on this transport protocol we present a dynamic media streaming architecture with near-optimal bandwidth utilization under equation-based congestion control. We evaluate PRRT's bandwidth utilization and compare it to recent media streaming standards such as HTTP-DASH (Dynamic Adaptive Streaming over HTTP).