Scalable Group Composition with End-to-end Delivery Semantics

This paper examines the problem of building scalable, fault-tolerant distributed systems from collections of communicating process groups while maintaining well-deened end-to-end delivery semantics. Our approach to inter-group communication takes advantage of modular group composition. We introduce a generic model which provides a distinction between intra-group and inter-group communication, and allows for independent modular composition of protocols for each type of communication. Using this model, we provide a framework for analyzing the end-to-end semantic behavior of messages, making it possible to determine exactly what delivery semantics a set of receivers will observe for a given message, even when those receivers are members of diierent process groups which use diierent group multicast protocols and enforce diierent delivery semantics. To realize this model, we present an implementation architecture which is based on a simple abstraction: inter-group routers. Inter-group routers provide a mechanism for the exchange of messages between process groups. They allow groups to communicate in a fault-tolerant manner without sharing group or message state information and without modifying the underlying communication protocols. Finally, we present simulation results of our architecture for a variety of inter-group topologies and communication patterns. Using the published performance data of an existing group communication protocol, we simulate the performance of our architecture to verify its scalability. We also connrm that the end-to-end semantics predicted by our model are actually enforced by the architecture.