Adaptive control of media applications in best-effort networks using reliable transport protocols

Abstract

Quality of Service (QoS) in continuous media applications is an area of current interest because of the increased delivery of stored and live media over packet-switched computer networks. Media applications are transported using the User Datagram Protocol (UDP) and tend to use a disproportionate amount of network bandwidth, which is considered to be unfair to applications generating Transport Control Protocol (TCP) traffic. Therefore, methods for application QoS guarantees that also aim at being friendly towards network utilization are desirable. The objective of this research is to improve the QoS delivered to end-users by half-duplex applications across best-effort packet-switched networks. This is accomplished using TCP at the transport layer, along with adaptive predictive control at the application layer. The proposed algorithms are rooted in traditional process control. End-to-end delay measurements and predictions are used to compensate for the dynamic effects of time-varying time delays. The effectiveness of the compensation methods is examined using a MATLAB-based packet-level emulator and the widely used network simulator ns-2. The results show that media delivery in best-effort packet-switched networks using reliable transport protocols, such as TCP, is feasible. Furthermore, media application control can improve the QoS perceived by end-users, if predictive control algorithms based on reasonably accurate end-to-end delay predictors are utilized. This research concludes that properly designed media application controllers can be made "network friendly," alleviating concerns related to the impact of such applications on network congestion levels.