Self-Adaptivity Towards Tomorrow's Heterogeneous Computing Systems

ICT adoption rate boomed during the last decades as well as the power consumption footprint that generates from those technologies. This footprint is expected to more than triple by 2020. Current High Performance systems are designed around the workloads that used to be common until some years ago. The advent of cloud technologies, on-demand task completion and the variety of different continuously changing workload crave for a redesign and better exploitation of the computational resources available nowadays. Until now, high performance systems have been designed to serve rather static workloads with high-performance requirements. The variety of application of the on-demand computing scenario is instead characterized by dynamic workloads, each having different performance requirements and criticality. A promising approach to address the challenges posed by this scenario is to better exploit specialized computing resources integrated in a heterogeneous system architecture by taking advantage of their individual characteristics to optimize the performance/energy trade-off for the overall system. Better exploitation although comes with higher complexity. This research thesis hence aims at addressing all the previous mentioned limitations, i.e. efficiency and dynamic workloads, by exploiting self-adaptivity to allow the system to autonomously decide which specialized computing resources are exploited to achieve a more efficient execution based on user-defined optimization goals, such as performance, energy and reliability while lowering the complexity for the end-user.