Over the past few years, the principal investigator has designed, implemented, and deployed a number of applications and application-level protocols that utilize IP Multicast for large-scale, multiparty remote-conferencing over the Internet Multicast Backbone or Mbone. In the current incarnation of the MBone tools, a separate application handles each media (e.g.,for audio,for video, and for whiteboard) and together these tools comprise a flexible multi-party communication system that scales to roughly 1000 participants. This NSF CAREER proposal builds upon this previous work on the MBone tools with a new emphasis on massive scalability. A principal goal is to design new end-to-end protocols complemented by new network services that will facilitate multi-party communication systems that scale to millions of hosts. At this extreme, ``multicast for the masses''becomes a reality. If successful,the traditional infrastructures for mainstream mass media ---e.g., analog cable networks and over the air radio broadcasts ---can be replaced by integrated packet-switched networks running highly scalable multicast protocols. The proposed approach for achieving this thousand-fold improvement in scalability is multifaceted. Initially, the investigator will design minimal but powerful extensions to the IP Multicast service model that will provide end hosts with more flexible mechanisms for exploiting the underlying network topology. In the current Internet architecture, network topology is deliberately hidden. While this has led to robust protocols and robust networks, it places fundamental scaling restrictions on end-to-end multicast protocols. Instead, the proposed work will investigate whether a properly modified IP Multicast service model can enable new end-to-end protocols that scale to the extreme degree proposed herein. One particularly promising extension to IP Multicast is a new forwarding service called subtree multicast or subcast. Subcasts provide the topol ogically-aware network localization that end-to-end protocols require in order to scale gracefully. Whereas a normal multicast packet is sent to every receiver in a target multicast group, a subcast packet is sent only to a well-defined, localized subset of the group. Leveraging subtree multicast, the proposed work aims to overcome many of the scaling limitations in current end-to-end multicast protocols, and in particular, seeks a new, subcast-based design for the local recovery component of modern reliable multicast transport protocols. But, more generally, subcasts hold the promise to have far-reaching impact on a number of end-to-end multicast protocols (e.g., announce/listen protocols like RTCP, session directory protocols, multicast address allocation schemes, congestion control probes as RLM, and distributed voting algorithms like SCUBA),and the proposed work will investigate the integration of subcasts into these existing protocols as well as new ones. Research prototypes based on subcast technology will be fielded in real, ``production'' environments and integrated to the extent possible into the proposed educational activities. To this end, the investigator will develop on-line curricula that utilizes an experimental webcast system to deliver course content over the network, for example, to students in their dormitories, homes, or offices. The resulting applications and tools will also be deployed in a ``distance-learning classroom'', currenty under development at U.C. Berkeley, which houses classes, seminars, and research group meetings. To realize this ambitious goal, the proposed project will integrate artifacts and infrastructure from the MASH multimedia networking project at U.C. Berkeley and other similar efforts in the network research community that are investigating and building new multicast applications and protocols.
