[ Homepage | Biography | Courses | Members | Publications | Research/Funding | Workshops/Presentations ]
Congestion Control for Continuous Media Streams - Source Rate Adaptation

Congestion control is the mechanism that enables a shared network to overcome the congestion, that is the state of sustained network overload when the demand for resources exceeds the available capacity. Internet's robustness and stability is due to the congestion control and avoidance algorithm implemented in its mainly employed protocol, Transport Control Protocol (TCP), which belongs in the class of the Additive Increase / Multiplicative Decrease (AI/MD) algorithms. TCP has been designed and successfully used for unicast reliable data transfer. However, its window-based congestion control scheme, which halves the transmission rate when losses occur affecting TCP flow's smoothness, and retransmission mechanism, which introduce typically end-to-end delays and delay variations, make TCP unsuitable for the continuous media (CM) streaming applications. The CM streaming applications require smooth flows' rate adaptation, low end-to-end delays and delays variation. Although, in contrast to data transferring services, they may tolerate packet losses, however the induced losses shall be controlled and kept low because high packet losses lead to deterioration of the perceptual quality of the CM stream, since the requirement for the timely delivery either prohibits the recovery of the lost packets through a retransmission scheme or the recovery mechanism fails to reach the decoding deadlines.

We are studying the issue of design congestion control algorithms for CM streaming applications. Our main results are (a) a self-adjusting rate adaptation algorithm for CM flows that presents improved fairness and smoothness in comparison to the same metrics under the basic AI/MD scheme, and (b) a multi-state algorithm that is capable of detecting and responding well to static due to access bandwidth limitations, i.e., ADSL users, as well as locking the rate of a CM flow to an appropriate value. The produced results have been recorded in the following publications: