Abstract
There is an increasing demand for using today's shared computer networks, such as the Internet, for group-based applications. Multicast is a developing communication technology designed to provide efficient multi-point message delivery for large-scale groups. Research directed at the data link and network layers has been very successful, and multicast service is now available in most best-effort networks. However, best-effort multicast networks do not offer quality of service guarantees such as bounded transmission delays and error rates. Therefore, group-based applications rely on multicast transport protocols for ordering, reliability, group management, and other end-to-end services.
This dissertation presents MESH: a novel, distributed transport protocol designed for large-scale multicast. We show that MESH's error recovery and receiver feedback service (1) achieves high application performance (i.e., low delivery latency and high throughput), (2) efficiently utilizes network and end-system resources, (3) provides a flexible error control model suitable for reliable, unreliable, and other error control paradigms, (4) provides timely state information required by congestion, flow, group management, reliability, and other control protocols, and (5) scales to large receiver sets and wide-area heterogeneous networks. Using the MESH framework, we design and implement a reliable protocol (MESH-R) and a deadline-driven reliable protocol (MESH-M) in a high-fidelity simulation of SURAnet and vBNSnet. We show that the performance and overhead of MESH-R and MESH-M compares favorably to extant transport techniques (namely, centralized, tree-based, unstructured, and FEC-based schemes) for bulk-data distribution and continuous media applications.
