Master Thesis - Past Projects - Abstract

Finite Horizon Prediction of Computer System State and its Applications to Reconfigurable Component-Based Real-time Systems

ID: LiTH-IDA-Ex-05/082-SE

Real-time systems (RTS) have a large variety of applications and are becomming increasingly complex. For this reason, real-time software designers have begun to focus more on adopting component-based software development for real-time system development. Namely, component-based software development enables real-time systems to be developed out of pre-defined components with low development cost, short time-to-market and high degree of reconfigurability for their designs to shorten the development time of RTSs. Also, component-based RTSs can dynamically be reconfigured by adding, removing, or exchanging components with another during run-time. Dynamic reconfiguration of RTSs is often needed in response to errors and added requirements of the running applications. Perfoming the reconfiguration by stopping the running RTS, updating the software, and then restarting the application is not suitable for systems with high availability requirements. Therefore must the software be up dated during run-time, i.e., a dynamic reconfiguration of the system. However, dynamically reconfiguring a system may increase the execution time of the tasks, thus, degrading the performance of the system as dynamic reconfigure may result in violations of quality of service (QoS) requirements. Hence, when performing run-time reconfiguration is is important that the system fulfills the QoS requirements in terms of utilization and deadline miss ratio placed on the system by the system operator, both under and after the reconfiguration. In this thesis, QoS guarantees are maintained by predicting the behavior of the RTS under reconfiguration in terms of utilization and deadline miss ratio before the reconfiguration takes place. This way the reconfiguration is done only if the predicted behavior satisfies desired QoS levels, e.g., overshoot in deadline miss ratio. In this work, we developeda model of component-based RTS systems suitable for predicting the behavior of soft RTSs under reconfiguration. The model accurately predicts the behavior in terms of utilization and deadline miss ratio before, during, and after dynamic reconfiguration of the system. Experiments show that the accuracy of the prediction is good under a variety of situations. The experiments also show that the accuracy of the prediction is good for a sets of tasks with different properties, i.e., execution time, inter-arrival time, and number of tasks in the set.

Keywords: real-time, feedback control, component, CBSD, aspect, AOSD, prediction

Author(s): Torgny Andersson

Contact: Mehdi Amirijoo