(A) hierarchical congestion control framework for reliable multicast

Abstract

In spite of its efficient way of data distribution, multicast communication is inherently greedier in bandwidth than unicast communication. Furthermore, multicast flows insensitive to congestion are likely to cause simultaneous congestion collapse in many parts of the networks. Therefore, the design of a multicast congestion control scheme is very important. In the case of a unicast session, a window-based scheme as in TCP achieves multiple time scales of congestion control, i.e., dynamic control combined with quasi-static control. For reliable multicast, the sender needs to receive immediate feedback about the status of all receivers in order to respond quickly to congestion occurring at many parts of a multicast tree. However, because of the large number of receivers involved, feedback packets from the receivers should be consolidated by designated members of the multicast group to reduce the overload at the sender. There is additional delay with aggregating feedback information. Therefore, it is unfeasible to employ a simple window-based mechanism to achieve multiple time scales of congestion control in reliable multicast. In this thesis, we propose a hierarchical congestion control (HMCC) framework for reliable multicast and evaluate its performance with respect to scalability and fairness. The main objective of the proposed framework is to achieve multiple times scales of control while satisfying the scalability and fairness of multicast congestion control. We developed two schemes of the HMCC framework: HMCC-CMFC and HMCC-PTC. The high-level component, TCP-friendly Traffic Volume Regulation (TTVR), is common in both schemes. The high-level control uses an end-to-end window-based mechanism, and control actions take place in a medium-term scale. It regulates average transmission rates of the sender by determining the amount of data packets to be transmitted for a certain time interval. The low-level control part of the HMCC framework uses either hop-by-ho...