Abstract
After more than 25 years of active research, real-time scheduling theory has shown a transition from cyclical executive based infrastructure to a more flexible scheduling models, such as fixed-priority scheduling, dynamic-priority scheduling, soft real-time applications, feedback scheduling, to name a few. Nevertheless, just a few scheduling algorithms are available nowadays to implement real-time applications. For instance, almost every existing real-time operating system provides only POSIX-compliant fixed-priority scheduling , but not every real-time application requirement can be efficiently fulfilled using only fixed-priority scheduling. There are real-time systems constituted of hard and soft real-time tasks that could be better scheduled using dynamic-priority policies. Also, it's been shown that dynamic-priority scheduling allows a better utilization of system resources. 

Recently, a few proposals have been published to integrate more scheduling policies into a real-time operating system, offering an option to expand the real-time operating systems scheduling capabilities, and to implement, prototype and test many theoretical real-time scheduling research work. Some of the authors propose that the new scheduling services be implemented as user-level schedulers, avoiding the need of modifying the operating system kernel.

From the related work published to date, the Application-Defined Scheduling model proposed by Mario Aldea and Michael Gonzalez-Harbour deserves special attention. They proposed an Application Program Interface (API) to create and use application-defined schedulers in a way compatible with the scheduling model defined in POSIX.

This thesis is about the study of the definition of user-level schedulers in real-time systems, and particularly discusses the POSIX-compatible Application-Defined Scheduling model, in order to identify problems not solved and to propose some extensions to the model. Furthermore, it proposes an implementation strategy of user-level scheduler to minimize overhead. The proposed model in this thesis simplifies real-time systems implementation, provides portability, and serves as a framework to develop and use user-level schedulers.