Doing Centralized Desktop Virtualization Right
There are two main centralized desktop virtualization mod-
els. The first, stateful virtual desktops, focuses on virtual desktops that are tied to each specific user. In this model, each user connects to his or her own particular desktop virtu- al machine(s). The VMs are stored in a central shared storage container, much as they are when you virtualize server soft- ware. Host servers running production hypervisors manage all of the desktop virtual machines and make sure they are highly available. This model often tends to require a signifi- cant amount of storage since each VM can easily take up dozens of megabytes or more. Users rely on an RDC link to connect to their VM PCs (see Figure 3).
The second model, stateless virtual desktops, focuses on the generation of virtual machines on an as needed basis. Machines can either be generated when the user connects or they can be pre-generated and linked to a user when a con- nection request occurs. The advantage of this model is that the machines are completely volatile and built on the fly. The core desktop image is generated and then, when the user is identified during the connection, the applications they require are applied to the desktop image (see Figure 4). The
user's data and preferences are also applied at logon. While you may think that this process is time-consuming and can cause user dissatisfaction, it is not actually the case. The ideal volatile desktop will also rely on Application Virtualization to profile applications only when the user actually requests them. And, because everything occurs on a backend storage area network, applications and user profiles are provided through high speed disk-to-disk exchanges which are practically transparent to users, even multiple users.