Emergent technologies in computing and telecommunication enable very high speed networking which motivate the development of a new category of complex applications. Example of such applications can be found in control systems, telecommunication management and energy management. These applications systems interact with a large and distributed environment. Complex applications have to take decisions based on parameters that may dynamically change in a non deterministic way and that represent the environment behavior. The overall development cost of these applications is prohibitive since such applications are large scale ones and must integrate different problem solving techniques and technologie. Reusable techniques are thus needed to develop such aplications.
We present a framework based on a library of components that participate to the development of such applications. A distributed architecture is defined. It is made up of three category of agents: environment agents representing the environment behavior, complex agents to process environment events, and supervisor agents to delegate the events processing between complex agents.
The library of the developped framework integrates a list of scheduling techniques since they represent a significant impact on complex applications performances. Depending on the environment\'s behavior, the most suitable scheduling technique is selected and integrated with the architecture of complex agents. Our approach is based on an experimental phase to generate the asssociations between the environment behavior and scheduling techniques. A rule base is thus generated from this phase. During the application system run time, each new environment behavior triggers the generated rule base in order to select and plug the suitable scheduling technique.'