Challenges and New Solutions for Live Migration of Virtual Machines in Cloud Computing Environments

Live Virtual Machine (VM) migration aims to move a VM between hosts without interruption to the services running in the VM. It is a cornerstone technology for cloud management and is critical for decreasing the operating cost of cloud providers and improving service quality. Many efforts have been made to improve the performance of live VM migration and a variety of achievements have been gained. But some key problems still require solutions or improvements. In addition, with the development and evolution of cloud computing, more and more applications are running in a data center, and some new cloud computing paradigms (e.g., Mobile Edge Computing (MEC)) are being developed as well. This brings more new problems and meanwhile new optimization opportunities for live VM migration. The goal of this dissertation is to identify and qualify the performance bottleneck, and then optimizes the performance of live VM migration in different deployment scenarios. Firstly, a comprehensive review of VM migration and optimization technologies is conducted. Then, according to the lessons learned from the review, a series of optimizations are proposed to solve different problems with live VM migration. For the migration in a Local Area Network (LAN) or within a data center, we build corresponding performance models, and design a new performance control algorithm which takes users’ requirements into consideration. This algorithm also solves the convergence problem with pre-copy migration. For the migration over a Wide Area Network (WAN) or across data centers, several mechanisms are proposed to improve the migration performance of VM storage data which is the bottleneck for across-data-center VM migration. Specifically, a three-layer VM image structure and a central base image repository are designed to increase data sharing between VMs and data centers. Based on these two structures, several optimization strategies are proposed for data deduplication and Peer-to-Peer (P2P) file sharing to further improve VM storage data migration performance. For the migration in MEC, several algorithms are designed to mitigate the problems with user mobility-induced VM migration. According to the presented migration environments, user mobility is divided into two categories. Based on this classification, two algorithms are designed to improve migration performance, two mechanisms are designed to reduce the service degradation resulting from VM migration and two algorithms are designed to lower the network overhead of constant VM migration due to user mobility