Efficient Workload Adaptive Scheme forLow-power Real-time Systems

Low-power schemes for real-time systemsthat dynamically vary CPU speed and voltage dependingon the probability distribution of the task’s work requirementwere implemented in previous studies to save energy.However, they did not consider the interrelationshipof the probability distribution between running tasks’work requirement to further reduce the power consumption.This paper presents a new low-power DVS schemeto combine all running tasks’ scheduling into a single optimizationproblem such that the frequency schedule of eachtask is determined by the system workload. The computationof the frequency assignment is optimized based on theprobability distribution of the workload’s clock demands.This DVS scheme keeps the real-time system schedulablebecause the longest time duration to execute these clockcycles is restricted to meet all tasks’ deadlines. Simulationresults show that it effectively reduces at least 30% lessenergy than previous studies without any negative effecton performance.