The Research And Implementation Of Virtual Machine Live Migration In Cloud-Based Network Operating System

With the rise of the network operating system based on the cloud computing and its applications in recent years, VM (virtual machine) live migration attracted the attentions both from the industry and academia and has been intensively studied as an efficient tool for system resource management and scheduling. The cloud-based network operating system is a brand new compute mode that the system resources are allocation through the network. VM live migration allows system to transfer a running VM from the physical host node to another one seamlessly and the whole migration process is completely transparent and unperceived to the user. Comparing with the conventional operating system, the network operating system offers many new functions with assistance of the VM migration, such as dynamic resource management, online system maintenance and upgrading, system-level fault tolerance and load balancing, and so on. As a result, the VM live migration improved system performance in resource utilization and security. In addition, live migration is the key bearing technology for green computing.This paper consists of three parts. In the first part, the subject background and research actuality are introduced in detail. Besides, we also study the principle and actual solutions of the virtual ization and VM migration for preparing for my own research. In the fallowing part, we proposed a load-based controlling scheme of VM live migration and design its implementation model. Especially, A VM memory migration method based on zero-copy is presented with supporting by the remote direct memory access. This proposed method avoids the migration data to be copied back and forth among the memory, the process cache and the network controller cache. In the last part, we established experiment plats to evaluate the performance of the proposed solutions in this paper. Known from analyses of experiments, the load-based controlling scheme proposed not only efficiency avoids the unnecessary VM migration, but also can ensure the load balancing between nodes in the system after migration. In additional, the experiment result shows that the proposed memory migration method shortens the total migration time and the downtime, and also reduces the impact on the host system caused by the migration.At the end of this paper, we summarize the work done in this article. Besides, targeted recommendations and the outlook of the future work are also presented at the same time.