Providing e-Transaction Guarantees in Asynchronous Systems with Inaccurate Failure Detection

In this paper we address reliability issues in Web-based transactional systems. We are interested in the category of systems characterized by stateless application servers. For these systems, a framework called e-Transaction has been recently proposed, which specifies a set of desirable end-to-end reliability guarantees. Within this framework we propose an innovative distributed protocol providing those reliability guarantees in the general case of multiple, autonomous back-end databases (typical of scenarios with multiple parties involved within a same business process). Compared to existing proposals coping with the e-Transaction framework, our protocol adopts a weaker approach to failure detection, i.e. it does not rely on any assumption on the accuracy of failure detection. Hence it reveals suited for a wider class of distributed systems, including those systems where the level of asynchrony makes stronger approaches to failure detection not feasible in practice. To achieve such a target, our protocol exploits an innovative scheme for distributed transaction management (based on ad-hoc demarcation and concurrency control mechanisms), which we introduce in this paper. We also provide hints on the protocol integration with conventional systems (e.g. database systems)