| 摘要: |
| 在数据中心的运营中运营商需要考虑如何在利润最大化的同时降低碳排放和提升服务质量,这些目标之间的平衡是一个巨大挑战.针对该问题,建立分布式数据中心负载调度的多目标优化模型,提出一种改进拥挤距离和自适应交叉变异的非支配排序遗传算法(ICDA-NSGA-II).在NSGA-II算法的基础上,通过对拥挤距离的改进能够提高算法的开采和勘探能力,引入正态分布交叉(NDX)算子和自适应变异算子增强种群的多样性,从而保证算法能快速、准确地得到Pareto解集.为了显示改进算法的有效性,对基准测试函数进行求解,仿真结果表明,改进算法相比于典型的NSGA-II和MOEA/D具有更快的收敛速度和精度,在分布式数据中心负载调度优化中,能够快速有效地给出满足利润、碳排放和服务质量等目标的Pareto最优解. |
| 关键词: 数据中心 负载调度 多目标优化 拥挤距离 自适应变异 |
| DOI：10.13195/j.kzyjc.2019.0702 |
| 分类号:TP273 |
| 基金项目:国家自然科学基金项目(62073300,U1911205)；数字制造装备与技术国家重点实验室一般项目(DMETKF2019018). |
|
| Multi-objective optimization of energy and performance management in distributed data centers |
|
HU Cheng-yu,YU Guo,YAN Xue-song,GONG Wen-yin,CAI Jun-yi
|
|
(School of Computer Science,China University of Geosciences(Wuhan),Wuhan430074,China)
|
| Abstract: |
| In data center operations, operators need to consider how to maximize profits, reduce carbon emissions and improve service quality. However, the balance between these objectives is a huge challenge, and in practical problems, we need get a group of solutions with good distribution quickly. Aiming at this problems, this paper establishes a multi-objective optimization model for distributed data centers energy and performance management, and proposes an improved adaptive mutation non-dominated sorting genetic algorithm (ICDA-NSGA-II) which improves the crowding distance and crossover operator. The crowding distance is improved in order to improve the dispersion and convergence speed of the algorithm based on the NSGA-II algorithm. Meanwhile, normal distribution crossover(NDX) operators and adaptive adjustment mutation operators are introduced to enhance the diversity of the population, so that the Pareto solution set can be obtained quickly and accurately. The experimental results on benchmark problems show that the improved algorithm has better convergence and distribution compared with the NSGA-II and the MOEA/D, and further results on the model of data centers show that the proposed algorithm can solve this problem quickly and accurately. |
| Key words: data center workload scheduling multi-objective optimization crowding distance adaptive mutation |
