Scalable Reliable Multicast Transport for the Internet

这是与南加州大学黛博拉·埃斯特林在NCR-9628729奖项下共同努力的结果。许多应用需要可靠的多点数据传输。一些众所周知的例子是网络新闻、电子邮件列表和网络路由协议，如BGP。然而，到目前为止，由于缺乏对IP组播和可靠组播传输的广泛支持，这些应用程序一直使用多个TCP连接来实现到多个目的地的可靠数据传输，这导致了较差的可扩展性和网络资源的低效利用。本研究的目标是开发一个框架和一套机制，以提供高效、健壮、可扩展和可靠的组播数据传输，既可以在非常大的网络上传输，也可以在非常大的组播组中传输。指导设计的基本原则包括应用层成帧(ALF)、IP多播组概念和接收者发起的组管理，以及TCP/IP的核心设计原则-简单数据报服务、可靠性的端到端方法以及对不断变化的操作环境的自动适应。该研究将(1)系统地研究所提出的自适应差错恢复方案的可伸缩性问题；(2)开发机制来定位差错恢复业务并大幅降低当前设计中的恢复开销；(3)开发用于多对多组播数据传输的流和拥塞算法；(4)研究新的会话消息分发模型以进一步提高可伸缩性。

This is a joint effort with Deborah Estrin, University of Southern California under award NCR-9628729. Many applications require reliable multipoint data delivery. Some well known examples are netnews, electronic mailing list, and network routing protocols such as BGP. Up until now, however, due to the lack of widely available support for both IP multicast and reliable multicast transport, these applications have been using multiple TCP connections to achieve reliable data delivery to multiple destinations, which results in both poor scalability and inefficient utilization of network resources. The goal of this research is to develop a framework and a set of mechanisms to provide efficient, robust, scalable, and reliable multicast data delivery both over very large networks and with very large multicast groups. The basic principles that have guided the design include the application level framing (ALF), the IP multicast group concept and receiver-initiated group management, and the core design principles of TCP/IP --- simple datagram service, the end-to-end approach to reliability, and automatic adaptation to changing operational environment. The research will (1) systematically investigate the scaling issues with the proposed adaptive error recovery scheme; (2) develop mechanisms to localize error recovery traffic and substantially reduce the recovery overhead in the current design; (3) develop flow and congestion algorithms for many-to-many multicast data delivery; and (4) investigate new models for session message distribution to further improve the scalability.