Date Published: February 6, 2019
Publisher: Public Library of Science
Author(s): Danqing Feng, Zhibo Wu, DeCheng Zuo, Zhan Zhang, Yong Wang.
To flexibly meet users’ demands in cloud computing, it is essential for providers to establish the efficient virtual mapping in datacenters. Accordingly, virtualization has become a key aspect of cloud computing. It is possible to consolidate resources based on the single objective of reducing energy consumption. However, it is challenging for the provider to consolidate resources efficiently based on a multiobjective optimization strategy. In this paper, we present a novel migration algorithm to consolidate resources adaptively using a two-level scheduling algorithm. First, we propose the grey relational analysis (GRA) and technique for order preference by similarity to the ideal solution (TOPSIS) policy to simultaneously determine the hotspots by the main selected factors, including the CPU and the memory. Second, a two-level hybrid heuristic algorithm is designed to consolidate resources in order to reduce costs and energy consumption, mainly depending on the PSO and ACO algorithms. The improved PSO can determine the migrating VMs quickly, and the proposed ACO can locate the positions. Extensive experiments demonstrate that the two-level scheduling algorithm performs the consolidation strategy efficiently during the dynamic allocation process.
Cloud computing is considered one of the most promising technologies to meet customer demand flexibly. Usually, it includes SaaS, PaaS, and IaaS. Software as a service (SaaS) provides access to complete applications as a service , and platform as a service (PaaS) provides a platform to develop other applications, such as the Google App Engine (GAE) . Infrastructure as a Service (IaaS) [3–4] provides an environment to deploy the managed virtual machines. A reasonable resource allocation strategy can help to consolidate resources and reduce energy consumption. From the perspective of the providers, the key issue to be solved is to maximize the utilization by reducing the fundamental costs. As a core technique, virtualization [5–7] provides an effective way to pack the application requests into the VMs. The virtualization technique can make full use of the utilization by decreasing the power consumption. Virtual mapping  has become one of the core techniques in datacenters, which provides a solution to the resource allocation. Generally, the problems to be solved are divided into two subproblems: when to migrate and where to locate.
In this section, we propose a two-level scheduling algorithm aimed at maximizing the utilization, avoiding SLA violations and reducing the energy consumption during the scheduling process. We divide the consolidating algorithm into three parts: the triggering part, selection part and location part. These parts are described in detail as follows. In the triggering part, we determine the hotspots by the score model. It describes the time at which the overloaded PMs migrate. In the selection part, we select the VMs from the overloaded and under-loaded PMs. Besides, we also quickly select the VMs from the hotspots by the PSO. In the location part, we place the migrated VMs into the selected positions by the improved Ant Colony Optimization (ACO) algorithm. The hybrid algorithm is shown as Algorithm 1. We provide a detailed description below.
Two types of experiments are designed in this paper. One is a simulated experiment, and the other is a set of real application request experiments. These experiments were implemented on the CloudStack platform to verify the validity of the proposed algorithm. The results demonstrate that the proposed algorithm improves not only the CPU utilization and the memory utilization and also reduces the SLA violation and energy consumption.
Traditional scheduling approaches focus on the energy model to reduce the overhead. However, additional factors have effects during the scheduling process. In this paper, we develop a novel consolidation algorithm that uses multiple objectives, such as minimizing the cost overheads [57–58] and the power consumption . First, we determine the hotspots by using the score model in the data center. When the score threshold reaches a specific value, the hotspots are identified. The score model solves the issue of when to migrate. Second, we quickly migrate the VMs by using the PSO algorithm. To save the energy overheads, we take the VMs in the under provisioning into the migrated list. This solves the question of which VMs should be migrated. Third, we propose an improved ACO algorithm that simultaneously attempts to minimize the rental cost and the power consumption. Using the Pareto efficiency leads to better quality in solving the resource consolidation problem. This solves the issue of where to migrate. We can then shut down the idle nodes and minimize the number of nodes. Finally, we evaluate the algorithm under simulated and real workloads. The results show that the proposed consolidation technique improves the utilization and enhances the scalability.