Research On Energy Efficiency Based On Resources Allocation For Data Center To Support Cloud Computing

Cloud computing has been widely used in different venders due to its advantages offered by the public, private and hybrid cloud, despite of many benefits to the cloud providers and cloud users, there are still several challenges faced in maintaining and managing the cloud. One of the key challenges is the inefficiency of energy consumption in cloud computing data centers, data centers consume so much power and some of these powers are not even used for resources computation.One of the main causes of energy inefficiency in cloud computing data centers is the inefficient of resources allocation. There are several factors influencing the improper resources allocation in cloud computing environments, so the resources allocation remains as a problem. Thus effective and efficient methods needed to be proposed to overcome this issue. In recent years, virtual machine migration method has been widely used as the way of utilizing the resources to the effectively, and the researches of effective virtual machines (VMs) migration policies and algorithms of resources allocation have been the hot topics.This thesis focuses on two main parts:the first part is to implement a proposed virtual machines selection policy for VM migration, which is known as the Threshold Maximum Utilization (ThrMaxU) policy. In this policy, we set fixed upper and lower host utilization threshold, dynamically in a specified time we check if upper and lower threshold are violated. When the utilization of a host is below the lower utilization threshold, all VMs in that host should be migrated and the host is shut down. In case when the utilization of some hosts are above the upper utilization threshold, we arrange the hosts violated the upper utilization in decreasing order, starting with the host which exceeded utilization portion most and VMs with the higher CPU utilization among of the hosts experiencing violation of upper or lower utilization threshold are selected first. We will adopt this selection policy and integrate it with the existing allocation policy called Modified Power Fit Decreasing (MBFD) to evaluate the power and performance trade-off. We use CloudSim toolkit to simulate our work and compare the performance of our proposed policy with other benchmarks in term of power consumption, percentage of SLA violation, number of hosts shutdown and number of VMs migrated during migration process. The simulation results show that our ThrMaxU policy can achieve less power consumption, less hosts shutdown, less VMs migrated during the migration process and the SLA violation is competitive with other policies.The second part is to propose a Modified Power Best Fit (MPBF) algorithm to allocate the resources globally; means in every allocation, the algorithm finds the server that provides less increase in power consumption and place the request there, so the total power consumption of a data center is minimized. The algorithm is designed based on the same resources computation capacity for the servers, means all servers used have the same capacity of receiving, executing and allocating the resources for users’ requests. We use MATLAB to simulate our MPBF algorithm and compare the performance with the benchmark algorithms in term of power consumption, number of resources used and the speed of execution of the user’s requests for different types of CPUs. The results of simulation show that our MPBF algorithm can achieve less power consumption increase and less number of servers used for computation processes for different type of CPUs, however the execution time of allocating user’s requests of MPBF takes2-3times compared to that of PBF algorithm and without DVFS method.