Design And Implementation Of A Private Desktop Cloud With Rapid Cloning Feature

Recently with the development of virtual desktop technology, the desktop cloud is used in various fields. It utilizes virtualization technology and remote protocol, to provide users with remote access to running virtual machines located in a centralized data center. It has a significant advantage in improving data security, reducing administrative complexity, using of resources effectively and flexibly. However, existed desktop virtualization products are expensive, and due to the large of the virtual machine images, it takes a long time in the case of large-scale cloning virtual machines.In order to solve the issue that it takes long time to clone the virtual machine image, the paper proposes a method that use the copy write technology to quickly clone mirror, and design and implement a private desktop cloud platform base on this. The implementation of the underlying platform is based on libvirt and KVM. Firstly, the paper designs the overall platform architecture diagram, and divides the platform into the service interface layer, management layer, public infrastructure layer and physical layer from the top down. Then base on the function, the platform is divided into physical machine management module, connection pool module, store module, virtual machine management module and scheduling modules, and then describe these modules in detail. In order to enable users quickly and easily access virtual desktops, the platform provides three remote access ways which are based on spice protocol, based on vnc protocol and based on web. Meanwhile, in order to improve the storage performance of the mirror, avoid it from becoming a bottleneck, this paper proposes that use moosefs distributed file system to store all the mirrors, so that the pressure of access the mirror will be shared by multiple nodes. In order to improve the scalability of the platform, this article designs a pluggable scheduling system, while achieving a stochastic scheduling algorithm and a scheduling algorithms based on load balancing. Finally, we test the function and performance of the entire platform, the test showed that the overall functionality and performance is good.