Optimum Utilization of Resources Through Restricted Virtual Machine Migration and Efficient VM Placement in Cloud Data Center

Introduction

Now-a-days cloud computing has become the fastest growing part of Information and Communication Technology (ICT) which provides an economical computing paradigm through its pay-as-you-go model. The flexibility to scale computing resources according to customers’ demands makes cloud computing much attractive. Large companies like Foursquare and Netflix (Amazon EC2, 2017) have moved their business services to Amazon Elastic Compute Cloud (EC2) which provides Infrastructure-As-A-Service (IAAS) cloud platform to their customers on rental basis. Similarly, many more companies are getting attracted towards cloud which provides a reliable IT solution at a lower cost. But the runtime performance of virtual machines (VM) in IAAS cloud is very much unpredictable due to resource contention of VMs (Armburst et al., 2010) To prevent this surplus amount of resources must be available in the data centers but it leads to inefficient resource utilization as many resources remain idle most of the time.

Studies have found that servers in many existing data centers are often severely underutilized due to over provisioning for the peak demand (Armburst et al., 2010). There is a large energy consumption gap between an idle server and an inactive server. Research shows that even completely idle server consumes almost 70% of the peak power (Fan, Weber, & Barroso, 2007). In 2013, U.S. data centers consumed an estimated 91 billion kilowatt-hours of electricity. Data center electricity consumption is projected to increase to roughly 140 billion kilowatt-hours annually by 2020 (Natural Resources Defense Council (NRDC), 2014). So, it is very necessary to prevent energy wastage for the enhancement of green computing and to reduce the operational cost of cloud computing.

So, the main challenge of a Cloud Service Provider (CSP) is to fulfill the Quality of Service (QoS) requirements at minimum energy consumption. It has to maintain the balance between energy consumption and performance degradation. This is popularly known as energy-performance tradeoff. To achieve this, we have to take care of two things: i) Hotspot mitigation and ii) VM consolidation which are described below:

• 1.

  Hotspot is defined as a condition when a host has inadequate resources to meet the performance demands (Mishra, Das, Kulkarni, & Sahoo, 2012). The reason behind hotspot is improper distribution of load which causes some of the nodes to be overloaded while some other remains underloaded. The overloaded nodes or the hotspots cause violation of SLA due to lack of resources. They also generate more heat which in turn increases the cost of cooling system and substantial emission of carbon-di-oxide (CO₂) contributing to greenhouse effect (Beloglazov & Buyya, 2010). So, to mitigate hotspot VMs are migrated from overloaded servers to other servers which are less utilized. It improves the performance but at the same time overhead is increased due to VM migration as it consumes the resources of both the sender and receiver machine.

• 2.

  Underutilized servers not only cause inefficient resource utilization but also increase the operational cost. To minimize this VMs are consolidated, i.e. VMs are migrated to fewer physical machines (PM) to reduce server sprawl. VM consolidation maximizes the number of inactive physical servers by consolidating the VMs on a minimum number of active physical servers. In this way it saves energy but since load is concentrated on lesser number of active servers, so chances of performance degradation will prevail. Moreover, it also causes migration of VMs from one server to another.

In this paper we have reduced the number of VM migration which takes place due to both of the above reasons. This not only saves energy but also increases the overall performance of the system by efficient utilization of resources. The contributions of this paper are as follows:

• 1.

  A probability function is defined based on the upper threshold of the server. This probability is used to decide whether VMs will be migrated from over-utilized nodes or not.

• 2.

  A probability function is defined based on the lower threshold of the server. This probability is used to decide whether VM consolidation will take place or not.

• 3.

  A novel method is proposed to find a suitable destination host for a migrating VM. The host is selected based on its current utilization of CPU, memory and network bandwidth.