计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (2): 74-80.

• 大数据与云计算 • 上一篇    下一篇

云环境下超启发式能耗感知调度算法

陈少淼1,李智勇1,杨  波1,2,李彦武3   

  1. 1.湖南大学 信息科学与工程学院,长沙 410082
    2.湖南财政经济学院 信息管理系,长沙 410205
    3.国家电网湖南省电力公司信息通信公司,长沙 410007
  • 出版日期:2016-01-15 发布日期:2016-01-28

Hyper-heuristic energy-aware scheduling algorithm on cloud computation system

CHEN Shaomiao1, LI Zhiyong1, YANG Bo1,2, LI Yanwu3   

  1. 1.College of Information Science and Engineering, Hunan University, Changsha 410082, China
    2.Department?of?Information?and?Management, Hunan?University?of?Finance?and?Economics, Changsha?410205, China
    3.State Grid Electric Power Company in Hunan Province Information Communication Company, Changsha?410007, China
  • Online:2016-01-15 Published:2016-01-28

摘要: 能耗感知调度的研究对云计算数据中心的可持续发展有着重要意义。能耗感知调度是一个NP难的多目标优化问题,目前云环境下的任务调度算法较少考虑能耗问题,且不能实现对能耗的灵活管理,随机搜索算法是一种解决该问题的有效途径,但其计算开销大,收敛速度慢。将异构云环境下的能耗感知调度问题定义为一个带约束的问题,即在一定的完成时间下优化系统能耗,以实现对能耗的灵活管理。此外,提出了基于在线学习的超启发式算法(OLHH),该算法结合电压调节技术,在设计了简单高效的启发式策略集的基础上,引进超启发式算法,并采用在线学习的方式跟踪启发式策略的表现,实现对启发式策略的合理管理,从而达到提高算法的收敛性能的目的。模拟实验表明,该算法能够实现系统能耗的灵活管理,且比传统的随机搜索算法有着更好的收敛性能。

关键词: 异构云, 电压调节, 能耗感知调度, 超启发式算法, 在线学习

Abstract: The research of energy-aware scheduling is essential to the sustainable development of cloud computing data center. The energy-aware scheduling is NP-Hard multi-objective problem. Most of task scheduling algorithms currently less take the energy consumption into account, and can’t manage the energy consumption flexibly. Stochastic search is a feasible method, however, this method has high computation overhead and the convergence speed is slow usually. In this paper, the energy-aware scheduling problem is treated as a constraint problem on heterogeneous cloud computing systems, i.e. performance constrained energy optimization, to manage the energy consumption flexibly. Besides, Dynamic Voltage Scaling (DVS) technique is used to build a simple efficient low-level heuristics set, and?on?this?basis it develops a Hyper-Heuristic scheduling algorithm based on On-line Learning (OLHH) to improve the convergence. The simulation experimental results show that the proposed method can manage the energy consumption flexibly and has an outstanding convergence performance than tradition stochastic search techniques.

Key words: heterogeneous cloud, dynamic voltage scaling, energy-aware scheduling, hyper-heuristics, online learning