Article Preview
TopIntroduction
Using the strong backbone of grid, cluster and utility computing, Cloud has gain lot of attraction and adoption in just less than a decade. Because of the pay-as-you-go approach and avoiding the hassle of large IT infrastructure investment and maintenance, the usage of Cloud has been flooded like anything, from individual users to corporate clients and in the domain of e-Commerce, Reality sector, Academic and what not. Mell and Grance (2011) defines Cloud as a “model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction.” The NIST further lists five essential characteristics of Cloud computing viz. (1) on-demand self-service, (2) broad network access, (3) resource pooling, (4) rapid elasticity or expansion, and (5) measured service. It also lists three service models viz. (a) Software as a Service (SaaS), (b) Platform as a Service (PaaS) and (c) Infrastructure as a Service (IaaS), and four deployment models viz. (1) private, (2) community, (3) public and (4) hybrid. These Cloud services and deployment models assist programmers having ground-breaking thoughts without huge capital investment in computing infrastructures to deploy their products in the real market.
The Cloud data centers usually comprise of a great number of well-configured and interconnected computing resources (Luo, Li, & Chen, 2014) which consume a significant amount of electricity for their functioning. Increase usage of Cloud computing has lead to augmentation in this electrical energy consumption by the huge amount of servers in a large number of data centers. According to estimation (Christian & Belady, 2007), infrastructure and energy costs would contribute about ¾th whereas IT would contribute just ¼th to the overall cost of operating a data center in next three years. Data show that the average energy consumption of a data center is comparable to that consumed by 25,000 domestic usages (Kaplan, Forrest, & Kindler, 2008). This has attracted consideration of research community in recent years. Out of many different mechanisms to address the issue, workload consolidation and task scheduling have been recognized as few of the popular techniques. Server consolidation works on the principle of minimizing active servers in a data center without compromising the performance of tasks. Sleep/Wakeup has been identified as one of the top classifications by (Brienza et al., 2016) in which some of the servers are switched off when not in use to save energy and are awakened whenever necessary because it has been seen that even idle servers consume about 70% of their peak power (Fan, Weber, & Barroso, 2007). In a nutshell, proper distribution of existing tasks among available servers may result into minimizing the active servers without negotiation Service Level Agreements (SLA) with Cloud users.
In this research, we wish to address the scenario by identifying overloaded host from the data center. For doing so, we study various existing statistical methods used by researchers and propose to use a novel method, Moving Range (MR), which has not been used by the researchers so far. Using MR method, both upper and lower threshold values for host utilization can be computed, which help us in detecting overloaded or underloaded host, respectively. Detection of overloaded hosts would result into reduction in SLA violation and proper balance of existing workload. We further wish to reduce the overall energy consumption of data center by effective load balancing across the available host.
Rest of the paper is organized as follows. In next section, we describes related work pertaining to detection of overloaded host followed by discussion on various statistical methods used by researchers to compute threshold values along with our proposal. In subsequent part of this paper, we illustrate the experimentation scenario and simulation results. At the end, we conclude our research followed by list of references used in the paper.