Energy-Efficient Resource Allocation in Green Clouds

Abstract

Cloud Computing is a business model that is being widely-adopted by enterprises and organizations. It has become the preferred computing platform for deploying applications and services owing to its inherent characteristics, including scalability, pay-per-use, rapid elasticity, cost saving, self-service, and broad network access. These characteristics of Cloud computing have played a major role in its widespread adoption. To accommodate the growing demand for computational resources, Cloud market players Amazon, Microsoft, Google, GoGrid, Flexiant, etc. have set up large sized Data Centers (DCs). These large scale DCs consume huge amount of energy. Further, with the proliferation of Cloud computing, more and more Cloud based applications with varying resource demands and diverse Quality of Service (QoS) requirements are coming up, which require dynamic allocation, reconfiguration, and reallocation of resources. All these requirements necessitate the development of energy-efficient and QoS aware resource allocation techniques that not only reduce energy consumption but also satisfy QoS requirements of the end users. Reducing energy consumption while providing QoS is the biggest challenge that the Cloud service providers are confronting with. Therefore, to achieve the set objectives of energy efficient resource allocation, an extensive literature review on resource allocation in Cloud computing has been done. The state of the art techniques in the area of power management and resource allocation in Clouds have been explored. The comprehensive study of energy-efficient resource allocation techniques in Clouds has been carried out to identify their inherent limitations. From the literature survey, it is apparent that the biggest challenge confronting Cloud service providers is related to energy consumption and QoS aware resource allocation. To address the energy and QoS related challenges of Cloud, a green Cloud framework, named ``SERVmegh'', has been proposed for efficient and robust management of resources. The framework considers various characteristics of the Cloud such as robustness, scalability, fault-tolerance, and energy efficiency. The proposed framework divides the entire functionality across the different layers and also provides well defined interfaces for communication between layers. The layered architecture bears negligible overhead in terms of communication cost. A comparative analysis of the proposed framework with various open source as well commercial Cloud frameworks has been done. Further, a resource wastage reduction based allocation technique has been proposed for reducing energy consumption. The performance analysis of the proposed technique has been carried out in simulated as well as OpenNebula based private Cloud environment. The comparative analysis of the proposed technique with state of the art resource allocation technique has shown up to 18% reduction in energy consumption. To enhance the utility of ``SERVmegh'' cloud, a resource allocation technique based on Ant Colony Optimization (ACO) has been proposed and implemented. The resources have been allocated to the jobs to minimize total cost of execution, total execution time and total energy consumption while satisfying QoS requirements of the end users. Each QoS parameter of a job has been assigned some weight value that indicates its priority over the others. ACO has been applied at two levels for efficient allocation of resources. First level ACO allocates jobs to Virtual Machines (VMs), whereas second level ACO allocates VMs to Physical Machines (PMs). Server consolidation and dynamic performance scaling has been employed to conserve energy. The effectiveness of this approach is evaluated in CloudSim with jobs having different resource demands and QoS requirements. Approximately 10\% improvement in energy consumption has been observed from comparative analysis with existing resource allocation techniques. Further, an improved energy and QoS aware resource allocation technique ``EQUAL'' has also been proposed and implemented. EQUAL is based on Antlion optimization for efficient allocation of resources to an application encapsulated in a Virtual Machine(VM). It can be operated in three modes, namely power aware mode, performance aware mode, and balanced mode. When operated in power aware mode, it strives minimize number of PMs in order to reduce energy consumption. In performance mode, the applications has been allocated to servers that have maximum available capacity at disposal. Whereas, in balanced mode, power and performance have been given equal weightage while allocating resources to the applications. The mode of operation can be changed by varying the values of the control variables. The proposed approach is implemented in CloudSim simulator for performance evaluation and validation. The experimental results show that EQUAL saves energy upto 15\% and improves QoS in terms of reduction in percentage of tasks missed their deadlines. The major issue with EQUAL is, that it can not switch between different modes automatically. The mode of operation has to be selected offline. This issue has been addressed by extending it to support self-optimization of resources. The proposed self-optimizing resource allocation system can automatically switch to power mode, performance mode, or balanced mode in accordance with the workload of the application. The necessary changes are made in the control variables to shift the resource allocation system to different mode of operation. For autonomic mode switching, resource utilization history of the applications is used for predicting their resource needs in the near future. The self-optimization of resources allocated to an applications has been carried out in four phases: Monitor, Analyze}, Plan and Execute. The VMs are migrated to offload heavily loaded machines or consolidate lightly loaded machines to fewer physical machines to save energy by switching idle machines to low power modes. Server selection for a VM has been carried out using Antlion optimization. The self-optimizing resource allocation system has been evaluated with CloudSim simulator. The results have shown improvement in energy-efficiency and QoS of the Clouds. The results show that the proposed energy-efficient and QoS-aware resource allocation technique efficaciously addresses the challenges of cloud.