Mixed-criticality Embedded systems (MCs) consolidate real-time tasks with different levels of criticality onto a shared hardware platform. One of a primary challenge in this domain is the integration of energy-aware techniques without violating the stringent timing guarantees of HI-criticality tasks. Many existing schedulers employ rigid strategies, such as permanently dropping LO-criticality tasks during a system mode switch. While this approach reduces workload, it does not necessarily reduce system-level energy consumption because processors in HI-mode typically operate at the maximum frequency across all cores in a shared DVFS domain. As a result, power remains high while the overall Quality of Service (QoS) is degraded. To address these limitations, this paper introduces DEW-WIN, a Dynamic Energy-aware Window-based scheduling framework for multi-core systems. DEW-WIN partitions time into fixed-size windows and adjusts the processor frequency for each window based on the energy profiles of ready tasks, enabling a fine-grained balance between performance and power usage. The algorithm\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'s effectiveness was evaluated through a hybrid approach, combining hardware implementation on an ARM Zynq 7020 board and detailed emulation using the gem5 simulator. The results demonstrate that DEW-WIN delivers significantly improved performance over a classic mixed-criticality scheduling method. In high-pressure scenarios, DEW-WIN achieved schedulability rates as high as 97.5%, whereas the classic method dropped to as low as 35%. Furthermore, DEW-WIN maintained system stability by remaining in the normal operating mode for over 90% of the execution time, showcasing superior system reliability and greater overall energy efficiency.

Keywords