High demand for computational power by business, science, and applications has led to the creation of large-scale data centers that consume enormous amounts of energy. This high energy consumption not only imposes a significant operating cost but also has a negative impact on the environment (greenhouse gas emissions). A promising solution to reduce the amount of energy used by data centers is the consolidation of virtual machines (VMs) that allows some hosts to enter low consuming sleep modes. Dynamic migration (replacement) of VMs between physical hosts is an effective strategy to achieve VM consolidation. Dynamic migration not only saves energy by migrating the VMs hosted by idle hosts but can also avoid hotspots by migrating VMs from over-utilized hosts. In this paper, we presented a new approach, called extended-placement by learning automata (EPBLA), based on learning automata for dynamic replacement of VMs in data centers to reduce power consumption. EPBLA consists of two parts (i) a linear reward penalty scheme which is a finite action-set learning automata that runs on each host to make a fully distributed VM placement considering CPU utilization as a metric to categorize the hosts, and (ii) a continuous action-set learning automata as a policy for selecting an underload host initiating the migration process. A real-world workload is used to evaluate the proposed method. Simulation results showed the efficiency of EPBLA in terms of reduction of energy consumption by 20% and 30% compared with PBLA and Firefly, respectively.

商业，科学和应用程序对计算能力的高要求导致创建了消耗大量能源的大型数据中心。这种高能耗不仅增加了可观的运营成本，而且还对环境（温室气体排放）产生了负面影响。减少数据中心能耗的一种有前途的解决方案是虚拟机（VM）的整合，它允许某些主机进入低消耗的睡眠模式。在物理主机之间动态迁移（替换）虚拟机是实现虚拟机整合的有效策略。动态迁移不仅可以通过迁移空闲主机托管的VM来节省能源，而且还可以通过从过度利用的主机迁移VM来避免出现热点。在本文中，我们提出了一种新方法，称为学习自动机扩展放置（EPBLA），基于学习自动机，用于动态替换数据中心中的VM，以降低功耗。EPBLA由两部分组成：（i）线性奖励惩罚方案，该方案是在每个主机上运行的有限动作集学习自动机，以将CPU利用率作为对主机进行分类的指标来进行完全分布式的VM放置；以及（ii）连续的动作集学习自动机作为选择启动迁移过程的负载不足主机的策略。实际工作量用于评估所提出的方法。仿真结果表明，与PBLA和Firefly相比，EPBLA的能耗降低了20％和30％。EPBLA由两部分组成：（i）线性奖励惩罚方案，该方案是在每个主机上运行的有限动作集学习自动机，以将CPU利用率作为对主机进行分类的指标来进行完全分布式的VM放置；以及（ii）连续动作集学习自动机作为选择启动迁移过程的负载不足主机的策略。实际工作量用于评估所提出的方法。仿真结果表明，与PBLA和Firefly相比，EPBLA的能耗降低了20％和30％。EPBLA由两部分组成：（i）线性奖励惩罚方案，该方案是在每个主机上运行的有限动作集学习自动机，以将CPU利用率作为对主机进行分类的指标来进行完全分布式的VM放置；以及（ii）连续动作集学习自动机作为选择启动迁移过程的负载不足主机的策略。实际工作量用于评估所提出的方法。仿真结果表明，与PBLA和Firefly相比，EPBLA的能耗降低了20％和30％。（ii）连续的动作集学习自动机，作为选择启动迁移过程的负载不足主机的策略。实际工作量用于评估所提出的方法。仿真结果表明，与PBLA和Firefly相比，EPBLA的能耗降低了20％和30％。（ii）连续的动作集学习自动机，作为选择启动迁移过程的负载不足主机的策略。实际工作量用于评估所提出的方法。仿真结果表明，与PBLA和Firefly相比，EPBLA的能耗降低了20％和30％。