VDI (or Virtual Desktop Infrastructure) has been all the rage in IT circles for the past couple of years. Once the virtualization of servers became the norm, looking to desktops as the next frontier in virtualization was natural. Unlike servers, however, desktops are affected by several major factors that need to be addressed and considered before we simply jump on the VDI bandwagon. VDI is a great and wonderful technology, but like any technology has its place and needs to be considered carefully.
First we have to look at an important concept that affects VDI – the idea of shared computing. With servers we naturally assume that the servers and the services that they provide are not for one to one consumption but will be utilized by multiple users. This concept exists and has always existed in the world of desktops and is often referred to as terminal services. Terminal Servers are the server world’s answer to the need for centralized desktop resources and have been standard since literally before desktops even existed.
It is an interesting bit of computing history that brings us to how the Windows world interacts with the concept of terminal servers. Most operating systems and nearly all that remain in use today are designed from the ground up, and always have been, as multi-user systems. The idea that one user would sit in front of the computer as the “main” user and others would either be second class citizens or not exist at all did not really exist and all users were considered equal. Windows, unlike the UNIX family for example, came from a world of single user systems originating with DOS and DOS/Windows which were not multi-user and when Windows NT was developed as a multiuser system a great many software developers treated it as they always had making software that would not function well or often even at all in a multiuser mode.
This software ecosystem that is unique to Windows (it could effectively never exist on Linux, for example, because such software would simply be seen as broken due to the way that the ecosystem and deployments exist) has created an interesting problem making some software and some tasks easily addressable by the use of terminal servers identically to how it would be tackled by any UNIX OS while many other applications cannot be addressed using a terminal server and require a dedicated operating system instance for every user.
It is because of this historical factor leading to a significant difference in software ecosystems that has created the fundamental need for VDI and explains why VDI arose as a unique phenomenon within the Windows world and remains, for all intents and purposes, exclusive to it. So, it is very important to understand, that VDI arose conceptually as a means of addressing a need that existed only by a deficiency in third party applications and not because of an intrinsic nature of Windows itself in its current form or because VDI was a better approach to virtualizing or hosting end user desktop services. In fact, we could even look at VDI as an unfortunate kludge only needed in situations where we want to virtualize or centralize desktop resources and where some of the software needed to run on those systems cannot run in a multiuser mode. VDI is a fallback mechanism for special situations and not a desired approach to virtualized or centralized end user services.
It is important to note that due to the widespread use of VDI and necessity of it that the investment into support technologies around VDI has led to the possibility that in many cases VDI might actually outperform terminal servers even though architecturally this seems almost impossible. Basically this is happening because the incredible amount of research and development going into the hypervisor layer may be outpacing the same components in the operating system itself making for potentially better CPU and memory management and resource sharing. This is completely dependent on the unique situation, of course, as every OS and every hypervisor and every set of VDI tools is unique as well as the workloads being tested so mileage will vary significantly.
Also of serious consideration is, because of the aforementioned Windows-centric nature of the VDI concept, licensing. If we were to look at VDI from the Linux perspective we would have little to no licensing concerns and VDI would be up against traditional terminal services based on technical merits alone, but this is effectively never the case. The biggest single factor around VDI decision making is Microsoft licensing.
VDI licensing is both expensive, as well as complex. Companies wanting to consider the virtualization of Windows desktop resources have to carefully way the benefits against both the large cost of appropriate licensing and also the potentially large overhead of license management. Moving into VDI will likely mean lots of expensive IT time dedicated to license research, monitoring and training which is an often overlooked aspect of licensing costs.
VDI is a somewhat difficult concept to speak about in generalities because it is a slightly amorphous topic. If we virtualize a desktop, does it not become a server? If we use an operating system intended for server use, does that change what is and is not VDI? Is VDI based around use cases, licensing or product categories?
The real answer lies in that to the industry VDI is technically one thing but in practical terms to Microsoft, the only key licensing player in the space, it means something somewhat different. VDI is technically the virtualization of one to one “graphical end user” instances – meaning a single virtual machine being used by a single user much as a traditional, physical desktop or laptop would be used. To Microsoft, whose concerns are slightly different than those of the industry, the term refers to the virtualization of Windows “desktop class” operating systems. If you virtualize Windows “server class” operating systems, Microsoft does not view you as doing VDI. So we have to understand these two views of the concept to keep from becoming confused. In fact, using Windows Server OSes to get around VDI licensing needs from Windows desktops has become very standard and common. However, we have to remember the kludge nature of VDI and while this does solve the failure to write software that is multiuser in nature it does not address the very real potential that software was written with expectations of desktop-branded operating systems and we are somewhat likely to find end user software that is either locked (intentionally or unintentionally) to desktop operating systems only or is potentially licensed only on those platforms.
The last major consideration around VDI decision making is that unlike servers which when virtualized are completely virtualized, a desktop cannot be treated in the same way because there is always a physical component to it. The end user will always need a monitor to look at, a keyboard to type on, speakers to listen to, and so on. So when we are looking to move to VDI we must take care not to overlook the fact that we are not eliminating the need to purchase and maintain desktops, we are simply moving where the operating system will reside. We may redeploy older hardware to be used for remote access, move to thin clients or the newly termed and mostly meaningless zero clients or use otherwise in use “fat clients” to pull double duty handling both activities as a remote access client as well as providing its own desktop services.
Certainly virtualizing the desktop offers us many great opportunities and much value if we are doing it for the right reasons and understanding the hows, whys and whens of VDI. It has sadly, like so many technology trends, become a knee-jerk reaction to want to move to VDI without performing proper evaluations and developing a clear picture as to how VDI will fit into our own environments. If we lack a clear reason for choosing VDI it will be very unlikely that we deploy it in a positive manner.
Finally it is very important that we consider the necessary skill sets that will be required in order to properly move to VDI. From a purely technical standpoint, throwing a Windows 10 VM onto Hyper-V constitutes VDI but from a practical perspective this is not how effective VDI will be designed. VDI not only requires the special licensing knowledge that I mentioned above but will typically involve rather unique knowledge of modern and very specialized VDI toolsets and products, shared storage as it applies to VDI, remote access protocols, thin clients or zero clients, and more. VDI deployments tend to be one of the technical and unique components of an infrastructure leading to a great number of unknowns and challenges for any organization.