Middleware-based data replication is a way to increase availability, fault tolerance and performance of networked information systems without modifying the underlying DBMS core code. However, if such middleware is not properly conceived, it will introduce an overhead leading to poor response times. In this paper we study the progression of solutions for problems typically entailed by replication, which led to the use of middleware for avoiding inconveniencies associated with earlier approaches. We survey several existing and projected architectures ?designed by our research groups? in terms of availability, consistency and fault toleran