Self-optimizing skeleton execution using events

This thesis presents a novel way to introduce self-configuration and self-optimization autonomic characteristics to Algorithmic Skeletons using Event-Driven Programming (EDP) techniques. Due to the use of event driven programming, the approach proposed in this thesis contributes to the current solutions as follows: • It is not related to a specific application architecture, like the component architecture. Thus, it has a broader scope, and is independent of the architecture used to implement skeletons. • It allows the introduction of structural changes during execution time. Therefore, it provides a higher level of adaptability compared with other solutions that only allows to introduce structural changes during compilation. • The estimates of expected future work can be calculated on-the-fly, and it is not limited to pre-calculated estimates. This thesis focuses mainly on guaranteeing a given execution time for a skeleton, by optimizing the number of threads allocated to execute an skeleton. The QoSs (Quality of Services) autonomically managed to be self-configured and self-optimized are: (1) Execution Wall Clock Time, and (2) Level of Parallelism. Other contributions are: • Design and implementation of a novel skeleton’s separation-of-concerns based on event driven programing. This novel approach allows the introduction of non-functional concerns to skeletons without lowering its higher-level programming. • Evaluation of different estimation strategies of future work was done. As result, a recommendation is presented to answer the question: Which strategy produces better estimates under which circumstances?.