A large volume of information is now accessible to
decision support systems within organizations and on the Internet. In this context,
managing heterogeneous and distributed data sources while respecting their autonomy has
become an important issue. This is why mediator systems have been developed. They provide
declarative access to a wide variety of heterogeneous and distributed data sources.
In this talk, we present the Disco mediator system that
has been developed in the context of Dyade, a R&D joint venture between Bull and
INRIA. The first part of the talk concerns general aspects of mediator systems. We give
examples of mediator based applications, discuss alternative architectures and present
existing systems. The second part of the talk focuses on Disco. We discuss the
architectural choices that have been made, we present the main research problems that have
been tackled and we describe some implementation aspects.