Reliability-Aware Energy Management in Mixed-Criticality Systems

Abstract

Mixed-criticality systems are introduced due to industrial interest to integrate different types of functionalities with varying importance into a common and shared computing platform. Low-energy consumption is vital in mixed-criticality systems due to their ever-increasing computation requirements and the fact that they are mostly supplied with batteries. We propose a novel reliability-aware energy management approach and three techniques, Monotonous-DVFS, Stretch, and Combined Monotonous-DVFS/Stretch in which energy management targets non-safety-critical functionalities. The Monotonous-DVFS technique lowers energy consumption by monotonously distributing slack times between low-criticality tasks while the Stretch technique lowers the energy consumption of mixed-criticality systems with the cost of degraded service in low-criticality tasks. Our Stretch technique extends both execution time and period of tasks while preserving their utilization. This leads to degrading the tasks’ service level due to a period extension that is exploited by Stretch for energy management. Experiments show that Combined Monotonous-DVFS/Stretch provides around 25 percent energy savings with only 5 percent service level degradation in low-criticality tasks in a heavily utilized system. The energy savings can be increased to around 56 percent with the cost of degrading low-criticality tasks’ service level to the minimum level, while preserving the original reliability of the system.