Adaptive application level QoS control for low-latency multi-object multimedia applications over best-effort IP networks

The lack of QoS support in today's best-effort IP networks poses a major challenge for low-latency multi-object multimedia applications (LMMAs). Fluctuating transport performance of the network infrastructure often leads to nonegligible packet losses and high delay jitter, which may significantly impair the integrity (including both the data and the temporal) of the presentation delivered. In this work, making only the very minimum of assumptions for the network infrastructure, we address this challenge and develop a generic passive application level QoS control scheme that is capable of dynamically balancing between multiple quality requirements and is able to ensure an overall satisfactory quality for end users.; We first investigate the objective performance measures reflecting the application level quality that end users perceive. We study and identify the principles regarding real-time dynamic balancing between the control of synchronization errors, which consist of synchronization phase distortion (SPD) and resultant Media Data Unit (MDU) losses, and the adjustment of the end-to-end latency for synchronization control using delay equalization. We then present an application-level QoS driven delay and synchronization control scheme for LMMAs. By making use of a distributed timing mechanism, the scheme enforces a closed-loop intra-stream SPD and MDU loss control, and can be configured flexibly with the choice of the inter-stream synchronization control mechanism. The scheme is able to track the sychronization errors and MDU loss in real-time and adaptively balance between the latency control and the synchronization and MDU loss control.; We detail how the proposed scheme operates given different transport channel behaviors. For networks that have stationary statistical properties, we present two approaches, using either heuristic or worst-case analysis, to obtain the clock adjustment amount in run-time. We take wireline AV conferencing applications as the example and investigate the resultant performance using a trace-driven approach. We show that the proposed scheme is able to adaptively control the resultant SPD and MDU loss performances by regulating the equalization delay. We further demonstrate through performance benchmarking that the proposed scheme has apparent advantages in terms of a tighter control on inter-stream SPD and a better capability of dynamically balancing between synchronization and MDU loss requirements and latency perference. (Abstract shortened by UMI.)...