Master Thesis - Past Projects - Abstract

Adaptive Control of QoS-Priority Partitioned Real-Time Systems

ID: LITH-IDA-EX-07/041-SE

Many computer applications of today offer customers services, which can vary in importance and quality. For example, different importance levels are connected to different prices with services like video streaming, which partly can tolerate low quality, i.e., lost information or delays in computation. Sometimes systems become overloaded and cannot process all the work assigned to them. The use of control theory can let the systems lower service quality where it is allowed and, thereby, execute as much work as possible. In this thesis we adress a problem discovered in a previous study, where the system under control was observed to change characteristics as the workload of the system changed. This leads to a controller, which was tuned for a certain workload, showing bad performance in terms of settling time for the controlled variable after transient overloads. The approach presented in this thesis is based on adaptive control theory where the system characteristics are estimated and controllers are tuned during run-time to give the system the desired performance independent of the workload. The results show an improved performance compared to a fixed-parameter controller where no adaptation of the controller is carried out. The improved performance of the adaptive approach was observed in halved settling times for the controlled variable with workloads that the fixed-parameter controller was not tuned for. Further a discussion is presented on the shortcomings of the evaluated approach and future work on this subject.

Keywords: Adaptive control, Quality of Service, Real-time system

Author(s): Per Brännström

Contact: Mehdi Amirijoo