Fine-grained layered multicast with STAIR
IEEE/ACM Transactions on Networking (TON)
| Hi-index | 0.00 |
IP multicast techniques were developed to provide a scalable architecture for the delivery of large volumes of content to multiple receivers. However, one of the significant challenges associated with multicast delivery to large audiences is providing a scalable congestion control mechanism that is not only compliant with the dominant Internet transport protocol, TCP, but also addresses heterogeneity in the network bandwidth across receivers since receivers may have different connection speeds and different congestion levels. In this dissertation, we propose scalable solutions for reliable IP multicast congestion control which accommodate heterogeneous network bandwidth across receivers and provide a fair share of network resources with competing TCP flows. Despite the considerable works in the area of multicast congestion control, these schemes have some major limitations; unfairness to TCP flows, non-scalability to large audiences, coarse-grained rate change, and high design complexity. We first demonstrate that fine-grained multicast congestion control protocol can be realized and implemented with reasonable costs and complexity. At a high level, our STAIR (Simulate TCP's Additive Increase/multiplicative decrease with Rate-based) multicast congestion control algorithm simulates the property of TCP congestion control scheme as closely as possible while imposing minimal load on the network and minimizing the amount of control traffic. We also propose another new approach to multiple rate congestion control that leverages proven single rate congestion control methods by orchestrating an ensemble of independently controlled single rate sessions in the design of multiple rate congestion control. The main advantages of our protocol called SMCC are simplicity, scalability, and modular design. In addition to our work on IP multicast, we also propose a new network architecture for reliable content distribution using an overlay network. There has been a significant amount of previous work in overlay multicast since unlike IP multicast, overlay multicast does not require additional network support in the current Internet. However existing approaches for providing both reliability and congestion control on overlay multicast impose fundamental performance limitations, such as dragging down all transfer rates in the system to the rate of the slowest receiver. The ROMA architecture we propose delivers a scalable solution that better accommodates a set of heterogeneous receivers.