Virtualized End-to-End Management Functions for Aggregated Control of Video Traffic Flows

The paper describes the management and control (M&C) functions of various network nodes in an end-to-end rate-adaptive video transport system. Mobile user devices download video clips by sharing the underlying network path from an ingress node. At the core software level, M&C functions realize the well-known AIMD (additive increase multiplicative decrease) based video rate control algorithm to handle congestion along the path. AIMD is exercised on the aggregated data flows at a source ingress node based on the 'loss reports' signaled from the receiver egress node. Our aggregated AIMD-based control reduces the signaling overhead, relative to the existing approaches that anchor an AIMD instance on each user device itself. This offers scalability, while improving the user-experienced QoS: such as low jitter in transfer rates and isolation against device faults. The offloading of aggregated AIMD-based control to the in-network overlay nodes also allows a reduction in the overall bandwidth usage. The software handling of 'last-mile' issues in the path between user devices and egress nodes (such as greedy users and access network channel sharing) are discussed, in a context of fine-granular video encoders in the devices. The paper also shows a virtualization of our M&C functions (as VNF modules) for deployment in large-scale video distribution networks --- such as YouTube.