Study And Design Of The Optimization Strategy Of Xen Virtual Machine Live Migration

In recent years, virtualization technology has become the leading study subject, and is widely used in the industry, gradually reorganizing the IT industry pattern. Virtualization technology isolates the hardware systems and software system to improve efficiency in the use of computing resources, saving operating costs, and enables the efficient of server to upgrade from20%-25%to80%; the memory use efficiency from the conventional35%to80%-40%.Meanwhile, with the application upgrade and resource integration, the demand of virtual machines live migration is rapidly increasing. Due to the migration mechanism is not perfect, the data migration process may result in downtime or data loss, migration timeout, thus affecting the business-critical data, applications and systems.Therefore, this paper focus on the open source Xen virtualization platform, and proposed optimization strategy of virtual machine live migration mechanism from bandwidth allocation and memory-intensive businesses perspective. Then the paper designed and implemented the prototype system based on the Xen virtual machine migration system architecture, which implements the live migration optimizing strategy, and can adaptivly select algorithm based on the application feature recognition model.First, the source code of the Xen bandwidth adaptive algorithm are in-depth analysis in this paper, followed by the comparative validation of the Xen virtual machine live migration mechanism performance under different migration scenarios. The results show that during the migration process of Xen, bandwidth resource allocation is not reasonable, pre-copy algorithm may fail when dealing with memory-intensive applications. Then, the paper proposed allocation of bandwidth optimization strategy, memory-intensive business applications optimization strategy, and business model for identifying business applications feature recognition. Finally established a prototype system composed by the data layer, algorithms layer and service identification modules. Experiments show that optimization effect of the system is obvious.