Article Preview
Top1. Introduction
The huge amounts of heat and carbon footprints are being produced in datacenter due to excessive internet usage, which may play a significant role in climate change. Therefore, environment-friendly computing must be required to promote green computing with the best service quality. In the current era, cloud services related to business, medical science, technology, entertainment, education, etc. are being used excessively throughout the globe. Smartphone applications, online gaming, video streaming, AI, machine learning, IoT based applications, etc. require processing resources on a large scale. Hence, millions of users may request online services simultaneously, which require vast infrastructure for datacenter and results in the proliferation of energy consumption on a large scale. Therefore, the management of EC in the datacenter with the best service quality is a challenging activity for cloud facilitators. These datacenters incorporate various types of resources, e.g., computing, storage, networks, etc. The allocation of these resources requires effective management to provide services to customers. Hence, dynamic consolidation of VMs is becoming a popular research activity to manage the VMs in the datacenter.
Virtualization is the key technique that enhances the efficiency of service quality of datacenter. When customers request computing services, cloud providers allocate virtual machines to them, which are based on the user requirement. But at a time, more than one user may ask for services that lead to heavy load on datacenter and affect hosts' performance. Some time hosts become overloaded, so there must be sufficient action to stabilize the load. Host overloading degrades the quality of service, which results in the penalty to the service provider. The load on hosts is based on VMs requirements, so; VMs may be transferred from one host to another for balancing the load in the datacenter. Such transfer of VMs is called migration in which VMs are reallocated from one host to another without interrupting the execution of the system (Bobroff, Kochut & Beaty, 2007). However, migration reduces the system's performance (Voorsluys, Broberg, Venugopal & Buyya, 2009). Hence, the migration should be less so that the quality of service (QoS) remains unaffected.
In the proposed work, Simple Moving Average (SMA) is applied for smoothing the utilization patterns. Further, Linear Regression (LinR) is used to predict the moving average for the next time interval. Initially, the SMA of CPU utilization of the host is taken, and then LinR is applied to predict the value of SMA for the next time frame. This predicted SMA is used further for the calculation of CPU utilization at the next time interval. In our approach, previous and current CPU utilization values of the host are used to determine the predicted CPU utilization, assuming that the predicted value, as estimated using LinR, is the average of previous, current, and predicted utilization. We applied LinR on the CPU utilization series which is generated using SMA for a specific time interval, and then utilization was predicted to detect the host overloading. The selection of VM from the overloaded host has been done using Mmt algorithm (Belograzov & Buyya, 2012), in the cloudsim simulator (Calheiros, Ranjan, Beloglazov, Rose & Buyya, 2011) using PlanetLab data (Park & Pai, 2006). Finally, VMs were placed onto the host with the lowest predicted utilization. Hence, we executed four combinations of host overloading, VM selection, and placement algorithm as LrMmt (Belograzov & Buyya, 2012), LrMmtPu, MaMmt, and MaMmtPu. SMA-LinR based approach (Ma) is the proposed work to detect overloaded hosts and estimate the predicted utilization. This predicted utilization (Pu) will be further used in the VM placement method. In order to achieve the appropriate solution for dynamic VM consolidation, the entire process begins with the prediction of utilization for detecting overloaded and underutilized hosts. All VMs are required for migrating from the underutilized host, and the suitable VMs are selected for migration from the overloaded host. Finally, a suitable host is selected for hosting VMs from overloaded hosts (Beloglazov & Buyya, 2012). The rest part of this article has been arranged in different sections; section II describes the research done in the related area. The problem formulation is done in section III. In section IV, the proposed approach and algorithms are described. In section V, the experimental environment has been discussed in brief. Experimental results are presented and discussed with performance analysis of the proposed work in section VI & VII and future work and the conclusion are given in the last section.