Flexible Fault Tolerance in Distributed Enterprise Communities

Abstract

We are witnessing the birth of the digital enterprise, in which many of the enterprise operations will be performed by independent software programs or by programs acting on behalf of humans. These heterogeneous agents are often managing different activities (such as stock trading or inventory management) in an autonomous manner and interact with each other in order to perform their job, creating distributed communities of agents. The involved agents are usually complex software entities, that should be able maintain a state that survives software or hardware failures. The classic solution is to use a relational database (such as Oracle), possibly replicated, to save the states of the agents in a manner that can survive failures and implement the communal invariants inside the database. However, keeping the invariants inside the database is difficult, since these invariants are application dependent and might be difficult to be formalized inside the database. But even more important, since the agents are distributed, keeping the states consistent in the presence of faults might require the use of distributed transactions, which are notoriously hard to handle and very expensive. In this paper we present a flexible, yet robust framework that will allow the states of the involved agents to survive hardware or software failures. In our system, the interactions of the agents that work in a given community are governed by a given law that is enforced in a distributed manner. The state of an agent is defined by the law and several options to handle the failures and keep the states consistent with each other are provided. The implementation of our system will employ the distributed coordination and control mechanism called Law-Governed Interaction} (LGI).