Virtual Machine Management for Efficient Cloud Data Centers with Applications to Big Data Analytics

Abstract

Infrastructure-as-a-Service (IaaS) cloud data centers offer computing resources in the form of virtual machine (VM) instances as a service over the Internet. This allows cloud users to lease and manage computing resources based on the pay-as-you-go model. In such a scenario, the cloud users run their applications on the most appropriate VM instances and pay for the actual resources that are used. To support the growing service demands of end users, cloud providers are now building an increasing number of large-scale IaaS cloud data centers, consisting of many thousands of heterogeneous servers. The ever increasing heterogeneity of both servers and VMs requires efficient management to balance the load in the data centers and, more importantly, to reduce the energy consumption due to underutilized physical servers. To achieve these goals, the key aspect is to eliminate inefficiencies while using computing resources. This dissertation investigates the VM management problem for efficient IaaS cloud data centers. In particular, it considers VM placement and VM consolidation to achieve effective load balancing and energy efficiency in cloud infrastructures. VM placement allows cloud providers to allocate a set of requested or migrating VMs onto physical servers with the goal to balance the load or minimize the number of active servers. While addressing the VM placement problem is important, VM consolidation is even more important to enable continuous reorganization of already-placed VMs on the least number of servers. It helps create idle servers during periods of low resource utilization by taking advantage of live VM migration provided by virtualization technologies. Energy consumption is then reduced by dynamically switching idle servers into a power saving state. As VM migrations and server switches consume additional energy, the frequency of VM migrations and server switches needs to be limited as well. This dissertation concludes with a sample application of distributed computing to big data analytics.