Article Preview
Top1. Introduction
Clouds are large pool of virtualized resources such as hardware (Processing Power, Memory, Storage and Bandwidth), development platforms and services that can be accessible to users through the network as per their demand. These virtualized resources are dynamically configured as per variable load with the help of scalability and load balancing. In order to satisfy growing need of customers for computing resources; large cloud data centres have been established across the globe containing thousands of servers such as Amazon, Google, Microsoft, IBM etc. to name a few.
Every data centre consumes huge amount of electrical energy for servers, network devices and cooling systems. There is a continuous rise of at least 100% power consumption from year 2010 to 2015 and at least 80% predicted rise from year 2015 to 2020 as per published report by Digital Power Group (Mills, 2013). Large data centres also affect environment due to carbon dioxide (CO2) emission. This high energy consumption may be due to inefficient use of computing resources in data centres. However, resource management is the major challenge for data centres because applications often experience variable workloads that demand dynamic resource usage. If resource demand of applications is not fulfilled, then application can face problems of higher response times or failures. Therefore, it is essential for data centres to establish Service Level Agreement (SLA) to ensure Quality of Service (QoS) to customers. Hence, data centres need efficient resource management techniques that deals with energy-performance trade-off by minimum energy consumption and SLA violation. Resource management techniques are based on server virtualization that need to address the resource management problem with two aspects. i) virtual machine (VM) placement and ii) virtual machine (VM) consolidation.
VM Placement is the process of mapping VMs to appropriate PM (Physical Machine) or host based on current demands of resources by VMs as shown in Figure 1 by resolving SLA between customers and cloud service provider. Let n be the number of PMs in data centre and m be the number of VMs to be placed. Number of possible mapping of m VMs onto n PMs is nm (He & Guo, 2011). Thus for a large values of n and m it is highly difficult, if not impossible, to manually explore all possible mapping and decide the best possible one. Automation of VM placement is therefore highly desirable.VM consolidation is a process of migrating VMs to some other PMs to satisfy their increased runtime demand of resources that could not otherwise be satisfied by their underlying PMs.
Resource optimization in data centre could be accomplished by making appropriate decisions about VM placements and consolidations during life cycle of VMs. Researchers have developed different VM placement and consolidation algorithms with trade-off between various conflicting performance parameters such as power consumption, SLA violation, number of VM migrations, Host Search Time etc.
This paper presents a novel approach called aiCloud that facilitates appropriate decisions about VM placements and consolidations.
Rest of the paper is organized as follows. Section 1 provides introduction while related work is presented in section 2. Section 3 deals with details of proposed aiCloud while experimental setup and results are discussed in section 4. Section 5 provides conclusions and future work.
Top2. Literature Review
Resource allocation during application deployment in data centre is modeled through mapping set of applications on to set of host servers with the help of virtualization. This problem has been studied by many researchers and proposed various solutions to achieve better values for performance parameters such as energy consumption, SLA violation, cost etc. Researchers have proposed solutions for efficient resource management by adopting several approaches like cost saving, performance, availability, traffic pattern, energy saving etc.