Will single app desktop VMs replace RDSH / Terminal Server in the future?

My belief is that as the world moves to devices and tablets and web apps and cloud whatever, the lowly Windows desktop application will be reduced to middleware, only used to remotely deliver old-style Windows applications that, for whatever reason, can't be migrated to native or web apps.

I've been writing about the future of the Windows desktop quite a bit recently. My belief is that as the world moves to devices and tablets and web apps and cloud whatever, the lowly Windows desktop application will be reduced to middleware, only used to remotely deliver old-style Windows applications that, for whatever reason, can't be migrated to native or web apps. When that happens we won't manage Windows desktop OSes on our endpoints. Instead we'll just deliver Windows apps as a service, most likely with a remoting protocol like HDX, PCoIP, or RemoteFX.

For years I just assumed the best way to do that was via RDSH / Terminal Server. (In fact way back in 2010 I wrote the article "The inverse bell curve of Terminal Server / SBC: This stuff is going to be huge again!") I figured we'd want to deliver these apps with RDSH because it would be cheaper that with VDI. But now I wonder—will that always be the case? Or in the future, will it be easier and cheaper for us to use single-app VDI instances to deliver our legacy Windows desktop applications?

How this will work

Just to make sure we're all on the same page, let's look at what I'm talking about.

My idea is that if you have a traditional Windows desktop application you'd like to deploy to your users without worrying about what kind of client device they have or without worrying about managing their local desktop, your only option is to run that application in the datacenter and to deliver it remotely via a remoting protocol. The actual platform the application runs on will have to be Windows (ignore WINE since it's not relavant here). So I'm making the assumption that the two choices are (1) a single session on a multi-session RDSH / Terminal Server host, or (2) a dedicated desktop (Windows 7/8) VM for the user that just runs that one application.

If you're going with the latter option, I'm assuming that you build a Windows 7/8 master image for each application you'd like to deploy, and then when a user connects the session broker spawns a new instance of the VM for that user. That VM only exists for that one user's session and is destroyed when he or she logs off. (While I'm a huge advocate of persistent desktops for scenarios when you're delivering the entire desktop to a user from the datacenter, if you're just delivering a single app then non-persistent desktops should work fine.)

In the past I assumed that the best way to do this was via RDSH / Terminal Server since that was the most efficient in terms of users per server (or dollars per user). But in 2013 you have to consider the improvements to the "VM per user" dynamics. All the hypervisors are getting really efficient at running lots of VMs (in terms of processor sharing, memory page sharing, etc.). And storage is pretty great nowadays too, with software-based single instance block-level storage like what we get from Atlantis or GreenBytes is almost a de facto best practice now.

Using a single instance desktop OS instead of RDSH / Terminal Server is also nice because it means that our remotely-delivered Windows applications are running on the same desktop OS that we're familiar with. And running a single application per VM image means that we don't have to worry about conflicts. So all the pieces are there.

I'm not trying to ignite the whole "VDI versus Terminal Services" debate. (Though it's been a few years since I talked about that. Maybe it's time to update it?) But in terms of delivering pure Windows applications from the datacenter, I feel like a few years ago it was no brainer to do it via RDSH / Terminal Services. But now? I'm not so sure.

What do you think?

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

I've been thinking about this for a while, too—the way that I've always thought about it was that Moore's Law will keep advancing, but the requirements to host and  remote a legacy Windows app (or a single app VM) will always remain the same during the life of the app (Unless there's some thing I'm missing here? But I'm just assuming that we're talking apps that don't evolve much...). And then presto...  VDI? RDSH? Who cares? In a few years that app will be just a tiny blip on the radar.


Using a desktop OS (under current Microsoft licensing) as the back end requires a VDA - which is often the biggest downside.  This scenario is supported by Citrix XenApp using the VM Hosted Apps deployment model.

It is a good option when some users occasionally need certain desktop OS only apps.  Usually if the majority of users need the apps all the time, that is an argument for full VDI.  A possible exception would be 'device-centric' users who won't accept the 'extra step' of launching their virtual desktop to get access to a desktop OS only app.


My math is still this.  90% of desktops can be handled in a XenApp/RDS style connection.  While only 10% would need the full Windows 7/8 OS.

It's about perception.  If you make the XenApp environment look like Windows 7/8, and all their apps work in that style.  Use it.  When polled most users thought they WERE running on a Windows 7/8 environment.

Why then have the Windows 7/8 VMs?  Applications that don't support Multi-User.  Systems requiring dedicated GPUs.  USB remoting of devices.  Integration with phone systems (should be under non-supporting multi-user, but important enough for it's own category.)

These numbers have been pretty solid in my real world implementations in the last 18 months.


I think a good question is 'why' do we deploy HSD instead of HVD.

I'm sure I will be corrected, but right now is it not because from a scalability and cost perspective HSD blows HVD away?

In 18/24 months time the incumbent hardware will probably be able to host at least 1000 sessions spread across several XenApp or RDSH server instances.  When this happens will you honestly want to lose this many sessions should a single server die?  The same hardware will probably support just enough HVD sessions to make 'scalability sense' and at this point would it not be prefereable to use HVD...when all said and done it's a lot less hassle to give every user their own dedicated desktop and resources than to adopt a shared model.

Does licensing make a play here?  What is the cost differential between hosting 500 users on Server2008/2012 as opposed to 500 virtual desktops?


I could see this happening.  In fact, I could see a number of applications pushed off to being a subscription based VM service from the application publisher.  Its only a matter of time, really.  Can you imagine Adobe just offering a very remotable VM-service to its customers.  No more costly desktops or laptops to maintain, and next to no deployment process.

I know it sounds like I'm suggesting thinClients are coming back, but I honestly believe there is potential there.  Just like there is a strong chance that application manufacturers could begin to sell prepackaged and ready for silent deploy versions of their applications, and in one stroke obviate the need for application packaging as a role.


While I find the concept of the single-app VM appealing for those special use cases where it makes sense, since they have been a feature in XA/XD, the use cases are still quite limited.  For MSP's, the SPLA restriction on desktop OS's would assumedly still apply, making them even less appealing in that scenario.  That said, I like @StevenOwen's concept of a remotable "VM-App-aaS" via subscription.  I think that could be a sensible evolution for those applications that would otherwise require a full desktop VM (or VM hosted app).  Of course at that point, you might as well put development resources into making a true, cross-platform SaaS application for greater compatibility.   I continue to wonder why something as simple as RemotePC still isn't considered for the same use cases where you'd want to do a single VM-hosted app?  I think that's one of the most underrated solutions to these types of issues and could be a solid, simple bridge strategy between traditional desktop applications and SaaS applications and their delivery.  


Can't loose site of the management overhead of managing more single app desktop sessions, which is a key benefit of XenApp. This problem however becomes smaller with the advent of Citrix Excalibur.

@markfermin I have ltd interest in remote PC to traditional PCs. I loose all the benefits of centralization that I want with RDS/VDI. That includes all the support data center infrastructure that I also want to centralize.

Also, moores law will benefit XenApp also. The risk of impact on large groups of users may be higher, but if it's one app, who cares about a small service interruption.

It all comes down to tiered app services IMO.

1) If you really care, a real desktop. I disagree with Brian that windows moves to middle tier as the PC form factor for things like multiple monitors, rich content creation, peripherals etc. will always work better running locally.

2) VDI will be used more to enable better centralization and consolidation of IT services. Becomes more true as technology advances continue to reduce costs.

3) RDS/XenAPP will continue to make sense when delivered from distributed back end services that are not close to a single VDI UI. It will also make sense for one off apps to large number of users for lowest possible TCO. MSPs will continue to drive this as a way to reduce the cost of everything.

4) I believe New cloud apps etc. will be consumed on companion devices like Tablets first. This will  mean that PC style apps stay around for even longer and over time apps begin to bridge multiple form factors with intelligent presentation layers.

5) I think single app VMs are therefore niche as app compat if 99.9% a bogus argument these days between Windows Server/Client.


Have you done any reviews of Atlantis, as looking at the architecture it seems to have a single point of failure built in.


What a hilarious string of comments, almost all agreeing that RDS is the way to go on the basis of scale and cost effectiveness.

Sure it is if you are comparing persistent VDI to RDS and stack them in the RDS like sardines, all sharing the same DOM Zero and the same core.  

Its like nobody has though through the security implications of hosting multiple tenants on the same infrastructure as most RDS shops do, nobody has thought through the implications of the fact that we still cannot properly separate distinct personalities securely on multi-tenant infrastructures securely.

No, we cannot and certainly not on a legacy terminal services stack from the last decade, despite your spluttering and assurances of security, you are all building your platform to fail any major kind of security and risk assessment, but you have it easy because your customers do not know any better.

I can see whats coming a mile away with RDS HSD providers stacking them high and selling them cheap, I know you fools can too even if you dont admit it.

You may get away with it too, none of you work with regulated industries who would never look at your service and none of your customers realise you are tricking them, you still lie to your customers about them having a Win7 desktop :)

More than half of you have simply repeated the second hand thought that RDS delivers higher user densities and cost effectiveness when it doesn't.

Keep on building your estates this way, it makes me happy that you guys have zero idea of the innovation underway in the DaaS space, that you all still feel you are in the right path with persistent RDS slices.

Me and mine are going to smash your businesses over the next 24 months, the time of RDS HSD is coming to an end no matter how many of you have made a living doing it for the last 30 years, no matter how badly your ecosystem wants to wring the last drop of revenue out of the tech before admitting its a lame mule.

When all you RDS monkeys retire, nobody will be left to repeat the same myths and the only customers you have will be those who do not know any better.