Energy-Efficient Scheduling for Dependent Tasks in DVFS-Enabled Heterogeneous Cloud Computing

Energy efficiency is a key challenge for large-scale data centers and cloud computing facilities. For applications with strict deadline constraints, typically modeled as Directed Acyclic Graphs (DAGs), meeting computational requirements while minimizing energy consumption on heterogeneous systems is a complex, NP-hard problem. Common approaches to reducing energy consumption include shutting down processors to cut static power and using Dynamic Voltage and Frequency Scaling (DVFS) to reduce dynamic power. However, static energy consumption cannot be overlooked in cloud computing, and focusing solely on either static or dynamic power does not lead to the minimization of overall system energy. To address this, we propose the DVFS-enabled Makespan and Energy Aware (DMEA) algorithm, which integrates deadline-awareness into scheduling decisions to minimize total energy consumption. The DMEA first sorts active processors by their energy consumption. It then uses a makespan-aware binary frequency search to reschedule the DAG, achieving the first round of energy optimization through global frequency scaling. Finally, the algorithm optimizes task slack times for local processor frequency scaling, yielding a second round of energy savings. Experimental results show that the DMEA outperforms existing energy-saving algorithms, achieving higher energy efficiency and deactivating more processors to further reduce energy consumption.