SLA and Performance Efficient Heuristics for Virtual Machines Placement in Cloud Data Centers

SLA and Performance Efficient Heuristics for Virtual Machines Placement in Cloud Data Centers

Oshin Sharma (Department of Computer Science and Engineering, Jaypee University of Information and Technology, Waknaghat, India) and Hemraj Saini (Department of Computer Science and Engineering, Jaypee University of Information and Technology, Waknaghat, India)
Copyright: © 2017 |Pages: 17
DOI: 10.4018/IJGHPC.2017070102
OnDemand PDF Download:
List Price: $37.50


Cloud computing has revolutionized the working models of IT industry and increasing the demand of cloud resources which further leads to increase in energy consumption of data centers. Virtual machines (VMs) are consolidated dynamically to reduce the number of host machines inside data centers by satisfying the customer's requirements and quality of services (QoS). Moreover, for using the services of cloud environment every cloud user has a service level agreement (SLA) that deals with energy and performance trade-offs. As, the excess of consolidation and migration may degrade the performance of system, therefore, this paper focuses the overall performance of the system instead of energy consumption during the consolidation process to maintain a trust level between cloud's users and providers. In addition, the paper proposed three different heuristics for virtual machine (VM) placement based on current and previous usage of resources. The proposed heuristics ensure a high level of service level agreements (SLA) and better performance of ESM metric in comparison to previous research.
Article Preview

1. Introduction

Cloud computing has become integral part of both industrial and academic communities due to its three main services, namely, Software as a Service (SaaS), Platform as a Service (PAAS) and Infrastructure as a Service (IAAS). These services can be used without any geographical restrictions (Arianyan et al., 2015). Continuous use of these services from their service providers such as Google, Amazon and Microsoft increases the usage of data centers. Therefore, high use of these data centers consumes huge amount of energy and raises an issue for efficient energy consumption. By using the capabilities of virtualization technology, this issue of an inefficient energy consumption can be resolved. Virtualization is the backbone of a cloud computing environment which enables the cloud user to use the resources of cloud data centers as a service. It lets the users to use the resources by dividing one physical machine into separate virtual machines, in which resources are in the form of logical or virtual. In the process of virtualization, virtual machine monitor (VMM) is responsible for virtualizing the hardware of host machine into virtual resources so that virtual machines can exclusively use them. Figure 1 shows the process of virtualization. Physical machine that host the virtual machine should provide all the resources that are required by VMs such as memory, bandwidth, storage and CPU utilization. Thus, by creating multiple VM instances on a single server, utilization of the resources can be improved. But on the other side, the virtualization is also risky because if one physical machine needs to be taken offline or breaks, several virtual machines will go down which further degrade the performance of system and degrade the quality of services mentioned in the SLA agreement signed between user and provider. This problem can be solved by setting redundant servers; if the first one i.e. primary server goes down, the second one i.e. secondary one will run the VMs until the first one is fixed. As, VMs are just data files therefore their migrations make it much quicker to recover if one of the server fails. Using live migration approach, VMs will be migrated to some other host without interrupting the current running applications, and this way live migration can improve the resource utilization of resources and switch off the idle host to minimize the energy consumption, but large numbers of system’s resources can be used during the process of live migration and may cause SLA violations. Therefore, there must be some proper policies to decrease the count of migrations during VM consolidation process. (Voorsluys et al., 2009) provides the impact of live migration on the system’s performance. Each phase of the process of VM consolidation presented by Buyya and can be improved to some extent to improve the performance of data centers. It is very essential for the cloud environment to deliver the reliable Quality of Services (QoS) mentioned in the Service Level Agreement (SLA) to cloud providers, so that cloud service providers can efficiently deal with energy-performance trade-offs.

Figure 1.


The focus of current work is to provide performance efficient dynamic consolidation or resource management strategies that can be applied inside data centers. Performance efficient consolidation means to reduce an unnecessary migration which leads to the degradation of performance of the system, to reduce SLA violations and to reduce the execution time taken for the overall consolidation process. For this, we have modeled the VM placement problem using classical bin packing problem, which is NP complete (Kou & Markowsky, 2007). Here we have proposed three different heuristics for VM placement based on the past utilization and available resources. These novel heuristics can be applied to the VMs and can provide a more precise solution to the problem. Our contributions in this paper are as follows:

  • Proposed three different heuristics for the selection of the host machine (VM placement) during VM consolidation process.

  • Evaluation of proposed heuristics in comparison to the benchmark algorithms such as PABFD, BFD and EPOBFD discussed in section 2.

Complete Article List

Search this Journal:
Open Access Articles: Forthcoming
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing