How User Virtualization Enables Desktop Virtualization
As desktop virtualization matures we are seeing that it will ultimately be delivered as either a client virtual desktop or a hosted virtual desktop (HVD). Most likely it will be a combination of both, and mixed in with more traditional physical desktops rather than as a complete replacement for the traditional PC. This article will focus on HVD as it is the only mainstream method of desktop virtualization available on the market today. However, the information provided here is just as relevant to client virtual desktops which we see becoming a reality in late 2010.
Today, there are (at a high level) two models with which to deploy HVDs. The first model is commonly known as “one-to-one,” the second as “one-to-many.”
For a quick overview, one-to-one is the method by which each user has their individual complete virtual machine (VM) running in the datacenter. This is considered as the full desktop (including all applications) being stored and executed centrally as opposed to running locally on a physical desktop or laptop/notebook device. This model has already been adopted across many organizations today by a means of a physical to virtual (P2V) conversion, such that a user’s existing physical desktop has been applied into the virtual world, and all previous physical operating systems (OSs) and associated applications are now running from the datacenter, virtually.
The second model is slightly more complex as it assumes that there is one central “gold image” from which all “user” virtual desktops are based. Each “user” then has some form of differencing disk that holds all changes from the initial gold image (specific to the current user environment), and together these items form the user’s personal working environment. The premise being that the enterprise only needs to hold a small number of these gold images that can be patched and managed centrally. This approach affords the enterprise the ability to not only lower administration costs of the desktop estate but also to lower the cost of delivering the desktops in the data center by considerably reducing the storage requirements. For instance, rather than requiring the enterprise to store the equivalent of the physical desktop hard disk in the data center for each and every user, the enterprise has one master image and then multiple (significantly smaller) differencing disks. An example to this might be as follows: 10,000 users, each of which have a laptop with 80 gigabytes of storage – in the one-to-one model there would be an immediate requirement for 10,000 x 80 = 800 terabytes of data center storage. On the other hand, in the one-to-many model the enterprise may be able to reduce the gold image to just 40 gigabytes (this is on the high side) that is shared, and then deliver a further 20 gigabytes for each differencing disk (again, this is on the high side). This example would give a total requirement of 200 terabytes plus 40 gigabytes for the central image. So, the one-to-many approach in this example results in a quarter of the storage requirements in the data center.
However, the one-to-one model is most prevalent, with almost all virtual desktop estates using this model today, in that each user has their own operating environment where the OS, application set and user profile information are welded together into one large virtual entity called the desktop. There may be application delivery technologies at play in building the desktop, but once it is built all three of these key components are firmly glued together. So, by moving the desktop into the data center, none of these key items are actually any better off. Sure they are now in a central place, so arguably are easier for administration teams to access for support purposes, but they are still together in the single entity.