A general scheduling framework for parallel execution environments

A major problem in the context of parallel and distributed programming is how to explore the resources of a multiprocessor architecture efficiently in order to obtain a gain in performance in the execution of a program. In the classical approach, to write a parallel program, the programmer needs, first, to detect the concurrent activities in the application and their respective synchronisations, and second, to distribute the activities and their data among the resources available on the architecture. In this approach we can find a low degree of program portability because the code will be dependent on the architecture resources. The article presents an alternative to this classical approach where the programmer is responsible only for describing its application by a graph of concurrent activities. It proposes a scheduling kernel that is independent of the program level to support the execution. This scheduling kernel was developed in order to support different load balancing schemes, improving program portability.