Autonomous Decentralized Collaborative Applications: an architecture for collaborative information-dissemination applications

In this paper we describe a model or architecture for autonomous and decentralized collaborative applications to support scientific research, collaborative learning projects and in general collaborative work. We first describe the application environment we had in mind: the design of a peer-to-peer collaborative newspaper application in a large network of schools. We have identified a generic component for decentralized information dissemination and we propose the Fast Consistency algorithm, a new weak consistency replication algorithm designed to satisfy the greatest global demand of fresh data in the shortest time, when network nodes only know some neighbours. This component may be useful to most collaborative or data intensive applications based on peer-to-peer or grid environments, requiring fast and selective replication of information on large and dynamic networks of autonomous agents with incomplete information, with clients or participants with varying degrees of interest, scattered or clustered in different locations, i.e. non-uniform geographic distributions of demand. We then describe LaCOLLA, a peer-to-peer architecture providing common collaborative support services. LaCOLLA supports three main abstractions (groups, events, group objects), services or mechanisms (event dissemination, group storage, group awareness, presence, object sharing, location, movement and replication), grouped in three classes of components (user agent, repository agent, group administration and presence agent) for peer-to-peer collaborative applications. It requires an infrastructure, which can be, among others, the Grid OGSA. This work has been validated by simulation of FC and LaCOLLA with several topologies, distributions of demand, and failures. The results confirm the validity, performance and self-organization of the system.