Scenario-based run-time adaptive Multi-Processor System-on-Chip

Modern embedded systems, which are more and more based on Multi-Processor System-on-Chip (MPSoC) architectures, increasingly require to be adaptive at run time to support complex and dynamic application workloads, dynamic Quality-of-Service management, etc. As one of the approaches for improving the adaptivity of MPSoC systems, dynamic application task (re-)mapping plays a crucial role in exploiting the system properties such that applications can meet their, often diverse, demands on performance and energy efficiency. The research of this thesis aims at improving these dynamic application mapping techniques to increase the efficiency of modern MPSoC systems by adaptively reconfiguring the system according to the dynamic behaviour of application workloads and the status of the target system. The application task (re-)mapping methods presented in this thesis belong to the class of hybrid task mapping approaches. They overcome the drawback of static task mapping techniques traditionally considered in embedded systems that are unable to support dynamic application behaviour as well as the drawback of pure dynamic (on-the-fly) task mapping techniques that typically only produce mappings of relatively low quality