The impact of Application Virtualization solutions in VDI, THE Reality Check

We (Sven Huisman, Jeroen van de Kamp and Ruben Spruijt) are pleased to announce the long awaited release of the Project Virtual Reality Check (VRC) Phase IV whitepaper. As promised: this one does a true deep dive on Application Virtualization in VDI.

We (Sven Huisman, Jeroen van de Kamp and Ruben Spruijt) are pleased to announce the long awaited release of the Project Virtual Reality Check (VRC) Phase IV whitepaper. As promised: this one does a true deep dive on Application Virtualization in VDI. We think we have interesting new content, and more importantly, many new insights. The primary goal of Project Virtual Reality Check phase IV is to investigate the overall scalability impact of Application Virtualization solutions within VDI. The three biggest and most commonly used vendors in the VDI market space are investigated: Citrix Application Streaming, Microsoft App-V and VMware ThinApp. These three solutions were compared in different streaming configurations.

It is currently almost impossible to conceive a VDI deployment without Application Virtualization. The management benefits of Application Virtualization are real and proven, especially within centrally hosted virtualized desktops. By isolating/virtualizing applications and streaming these on demand, a much more dynamic and flexible delivery of applications is possible in comparison to classic MSI installations. Typically, when a VDI deployment needs to support many applications, Application Virtualization becomes a key component to its long term success. Besides all the Application Virtualization benefits there are some down-sides as well. Some of these are: not all applications can be virtualized, interaction between applications can be a challenge and there is an overhead on the system resources such as CPU, Memory, storage.

Depending on the use case scenario, Project VRC discovered that this impact can be significant. Test results show that Application Virtualization has impact on the VDI user density, which can be decreased by 20% to 45% when Office 2007 is completely virtualized. This should be considered as a worst case scenario. When only a couple of specific (business) applications is virtualized, the session density decreases by only 3 to 12%. In practice the Application Virtualization overhead will be highly dependent on how often virtualized applications are started, and how much file IO and registry access they generate: specifically the creation of the virtualization ‘bubble’/environment for the application can have a significant overhead.

Both Microsoft App-V and VMware ThinApp are very close in overall performance overhead, resulting in very similar VSImax scores. However, Citrix Application Streaming overhead was considerably higher: the overhead is currently more than App-V and ThinApp. The highest VSImax score was achieved with ThinApp 4.6, where the ThinApp packages were locally stored in the VM image. From a disk I/O perspective VMware ThinApp showed the least overhead throughout the test. Interestingly, the response time of the file-open dialogue is much higher with Citrix Streaming, this clearly impacts VSImax results. Also, it is interesting to see how Application Virtualization is maturing, for instance ThinApp 4.6 clearly outperforms ThinApp 4.5 in our tests. The storage impact on read and write IO was also investigated and the general conclusion is that streaming applications will decrease the read IOs by 20% to 44% and increase the write IOs by 20% to 44%. From a management point of view, choosing for demand application streaming as the delivery method brings considerable management benefits and will offload read I/O’s. However, the impact on write I/O’s should not be neglected.

If you are looking for an independent advise and a ‘Reality Check’ in relation to virtualizing Terminal Server and Desktop (VDI) workloads, if you are curious about the impact of different hypervisors and the performance differences with various hardware, if you are searching for best practices for your virtual Desktops and if you are curious about the performance impact of different Application Virtualization solutions within VDI … the different Project VRC whitepapers are a must read. The ProjectVRC phase IV whitepaper "Application Virtualization Impact on VDI"is available for download here

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

While I like what VRC does and shares with the community, and am ok with login consultants using this as a marketing vehicle to upsell services, I have to do a little VRC reality check of my own.

"It is currently almost impossible to conceive a VDI deployment without Application Virtualization."

Come on guys, do you really believe that? You already know that the layer cake is not happening anytime soon, and most people will just default to XenApp for cost. You also know what most people just deploy VDI 1-1 using exactly the same bad/good management practices that they have today to get the benefit of centralization. You don't need application virtualization for that and in fact it just makes things more complex and expensive when the truth is you have to buy EA and MDOP from MS and App virtualization does not meet the need for all apps, so you end up with more overhead.

So good that you guys do some great work to show overhead etc, but to assert that app virt is required for VDI/desktop virtualization today is just not true.

Given your platform perhaps you could do more to highlight the BS. Ask the hard questions. i.e Why the F does MS not make App-V a free platform feature that the development community build on? Ask why Citrix and others are still in the business and have not built on top of App-V as a platform like they have for RDS? Is that because MS is a monopoly and wants to protect MDOP revenue and hence keep the system closed? Why is ThinApp important when you can't manage it? Sure run an exe from a mapped network drive is fine, but you call that an enterprise solution that is broad for the industry. May be it is, but I guess that's the consulting markup :-)

The real issue IMHO is that we are still talking about Windows apps. Did you hear that. Windows apps, i.e. running on a Microsoft operating system. I.E we are all MS B I T C H E S and they control the architecture. Toy solutions from Citrix and VMware that do unnatural things to the apps are never going to be for the masses, so please don't claim App virt is therefore strategic from these people. MS wants to protect their $$$$$ with EA and MDOP. App-V is a push from their sales organization to help with that. Nobody at MS cares about App-V as it's not a Windows platform component.

Therefore Application virtualization for the masses has already failed and I bet in 2014 when most people are on Windows 7, they still use MSI as the majority format with MS pushing Systems Center to manage it. That's the game and to marry app virtualization to desktop virtualization as a requirement is a mistake. It's an evolution that may never happen due to money concerns before customer solutions.


Having some sequencing trouble eh @appdetective? :)


Just to be able to compare properly -- what type of compression are you using in App-V ? By disabling compression in ThinApp, you are giving it an unfair advantage -- given that compression reduces the number of simultaneous sessions from 81 to 56 for ThinApp, this impact is significant. Did you disable compression also in App-V packages as a comparison?

As also with the previous VRC papers: how many runs are you using for the numbers? With numbers so close to each other (all within ranges of a few %), it seems very hard to make any conclusions based on only a few runs!

I think this is a very interesting project, but as long as the testing methodology does not improve, some precaution should be taken with the value of conclusions drawn from the data in Project VRC...


Hi TiJa,

All the test scenario's are executed more than three times.  

We have executed App-V with and without compression. There was a small performance impact with compression enabled.(default is disabled).

Thanks for your positive feedback,  when you have questions let us know.