If ubiquitously deployed, IP Multicast promises to provide an efficient datagram service for an arbitrary sending host to reach an arbitrary and dynamic set of destination hosts anywhere in the Internet. Unfortunately, two very difficult problems -- interdomain multicast routing and viable end-to-end multicast transport -- have yet to be solved and deployed satisfactorily.

This paper proposes a new multicast service model called the "Breadcrumb Forwarding Service" (BCFS) synthesized from the EXPRESS service model and the network component of the Pragmatic Multicast protocol (PGM). Like EXPRESS, BCFS utilizes explicit-source group join and like PGM, enhances the network forwarding architecture with finer-granularity group control. To demonstrate the flexibility and efficacy of BCFS, we developed a novel reliable multicast transport protocol, Rainbow, for bulk data transfer built on top of BCFS. In this approach, each receiver runs an independent, window-based congestion control algorithm modeled after TCP, thereby affording "TCP friendliness" while retaining the efficiency of IP Multicast. To enhance scalability and support asynchronous receiver behavior, we employ a Digital Fountain at the source. In this paper, we detail BCFS service model, describe how Rainbow builds on this service model and on the Digital Fountain abstraction, and evaluate the resulting system. Our simulation results show that Rainbow/BCFS performs well in a variety of configurations and is "TCP friendly".