An energy-aware scheduling method for parallel tasks based on an adaptive differential evolution algorithm in a multi-cloud environment

With the increasing data and computing scale, the energy consumption of computing is increasing greatly. This study focuses on the scheduling problem of parallel tasks in a cloud environment. Most of the work models tasks according to DAG (Directed Acyclic Graph) and selects a working state based on DVFS (Dynamic Voltage Frequency Scaling) to reduce energy consumption and meet other QoSs (Quality of Services). In contrast to these works, we focus on the task in which parallelism cannot be changed during execution, and the task model supports slot time in a heterogeneous environment. In the paper, we propose a SAEADE (An Self-adaption Differential Evolution Energy-Aware Algorithm) to schedule resources, which considers the parallelism of tasks, the selection of resources, and their working states simultaneously. SAEADE initializes the data by the sine function. During crossover and mutation operations, SAEADE selects the strategy by a roulette algorithm among the three methods: (1) DE-Rand, (2) DE-current-to-best, and DE-rand-to-best. Simulations show that SAEADE performs well in terms of makespan, energy consumption, the number of completed tasks, and the number of completed instructions. Compared to the performance of PSO (Particle Swarm Optimization), SAEADE also has good performance in efficiency.