An Application-Controlled Framework for Reliable Multicast Transport

IP multicast has enabled a number of large-scale collaborative applications, several of which require reliability. However, no single reliable multicast (RM) transport protocol fits the requirements of all classes of applications and functions efficiently over all types of heterogeneous networks. In this paper, we propose a novel framework approach to RM protocol design that involves a generic core protocol skeleton that can be customized in different ways resulting in protocols that are optimized for specific applications. We identify three key issues that must be tackled before such a flexible architecture can be realized. First, to accomodate diverse applications, the semantics of message delivery (i.e., reliability and ordering) must not be fixed, but rather must be flexibly controllable. Second, to scale to large and dynamic groups, our multicast transport protocol framework must be robust against host failures and group membership changes. Third, in combination with a “safe” congestion control scheme, our transport framework must allow applications to manage available bandwidth optimally. In this paper, we discuss our solutions to these problems. To demonstrate the power of our flexible framework approach, we have designed and implemented a prototype of our transport protocol library, called libsrm.