Footnote:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Description:
Multi- and many-core embedded systems are increasingly becoming the target for many applications that require high performance under varying conditions. A resulting challenge is the control, and reliable operation of such complex multiprocessing architectures under changes, e.g., high temperature and degradation. In mixed-criticality systems where many applications with varying criticalities are consolidated on the same execution platform, fundamental isolation requirements to guarantee non-interference of critical functions are crucially important. While Networks-on-Chip (NoCs) are the prevalent solution to provide scalable and efficient interconnects for the multiprocessing architectures, their associated energy consumption has immensely increased. Specifically, hard real-time NoCs must manifest limited energy consumption as thermal runaway in such a core shared resource jeopardizes the whole system guarantees. Thus, dynamic energy management of NoCs, as opposed to the related work static solutions, is highly necessary to save energy and decrease temperature, while preserving essential temporal requirements. In this thesis, we introduce a centralized management to provide energy-aware NoCs for hard real-time systems. The design relies on an energy control network, developed on top of an existing switch arbitration network to allow isolation between energy optimization and data transmission. The energy control layer includes local units called Power-Aware NoC controllers that dynamically optimize NoC energy depending on the global state and applications’ temporal requirements. Furthermore, to adapt to abnormal situations that might occur in the system due to degradation, we extend the concept of NoC energy control to include the entire system scope. That is, online resource management employing hierarchical control layers to treat system degradation (imminent core failures) is supported. The mechanism applies system reconfiguration that involves workload migration. For mixed-criticality systems, it allows ...