Is it finally time for Microsoft to buy Parallels?

One of the interesting things Peter Ghostine brought up on our joint webcast last week was that he still thinks Microsoft might build or buy Parallels Containers “OS virtualization” technology.

One of the interesting things Peter Ghostine brought up on our joint webcast last week was that he still thinks Microsoft might build or buy Parallels Containers “OS virtualization” technology. (Parallels Containers was previously known as Parallels Virtuozzo or SWsoft Virtuozzo.)

A lot of people have thought that Parallels Containers would be a great technology for Microsoft to own (myself included), although frankly I thought the window of opportunity had passed and the whole idea sort of slipped my mind.

But wow. If Peter is right and Microsoft does end up building or buying a Containers-like technology, that would be pretty huge.

What is Parallels Containers?

Parallels is a virtualization software company with a product called “Containers.” Containers’ OS virtualization is kind of like a combination between terminal server and a hypervisor.

Containers takes a single copy of Windows Server and divides it into a bunch of “containers” that each have their own machine name, IP address, services, accounts, etc. For all intents and purposes, each container looks and feels like a completely separate VM. But in reality only one copy of Windows is running. Each container is technically more like a sort of “super isolation” environment instead of a true VM.

In the real world this means that you can get VM-level isolation of sessions with server scalability more like terminal server (as opposed to a hypervisor running a bunch of VMs).

Why use Parallels Containers?

The biggest use of Parallels Containers in the real world is for all those Internet hosting companies that offer “private servers” for just a few dollars a month. They use Containers to put dozens of customers on the same physical host (all while each customer thinks they have their own private Windows server).

The other killer use case is for VDI. Since Parallels Containers allows you to get the isolation of VDI with the session density of terminal server, it’s truly the best of both worlds.

Probably the main reason that it’s not used for every VDI deployment is because the leading VDI vendors—VMware, Citrix, and Microsoft—all have their own hypervisors that aren’t Parallels. (I know for sure Quest, Ericom, and Leostream—three connection broker vendors who are not tied to one specific hypervisor—all support and love Parallels.)

What does this have to do with Microsoft?

For years Peter Ghostine has been saying something along the lines of “if Microsoft wants to dominate the VDI space and kill VMware in the desktop arena, all they need to do it buy Parallels and no one will ever look anywhere else!” (That’s not an exact quote but rather a paraphrase based on stuff he’s said.)

While I agreed in principle, I figured that if Microsoft actually wanted to do that, they would have by now. Unfortunately I thought my dreams were dashed when Microsoft announced their feature set for terminal server (now called Remote Desktop Session Host) for Server 2008 R2. Among the new features are per-session IP address virtualization, Microsoft Installer Service per-session virtualization, and per-session “fair share” CPU management. (In other words, they’re beefing up the per-session isolation on their own.) So I thought if Microsoft was really going to buy Parallels then they would have done it already and not spent the time or effort developing these new R2 features.

Then again, maybe instead of spending a billion dollars on Parallels, it’s cheaper for Microsoft to modify terminal server and incorporate App-V into their own home-grown super isolation for sessions hosted on terminal server?

What do you think? Could Parallels Containers dominate the space if Microsoft owned them? Can Microsoft build this on their own? Is it too late? Or is Parallels just too obscure to matter in any significant way?

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

Cool 4.5 RC supports Windows Server 2008, and you can have hyper-v running SxS.

I assume the containers can handle kernel level drivers in isolation?

I think containers are still relevant if you can get a very high client count going.


Actually, I think that probably the main reason everyone is not using it is because the Parallels VDI 'desktop' is a server OS, 2003 or 2008, rather than XP. From a technical perspective, the differences may be academic, but I believe the difference frightens of  many potential users. How easy would it be to get vendor support for desktop apps running in that environment? Probably not much more so than for a standard XP virtualized desktop, but it just feels like another layer of unnecessary complexity.


Parallels is not full session isolation. Try installing drivers per VM and see what happens. VM's give you full isolation and hence not a desktop experience. Try installing TS in a Parallels VM, I never got it to work. In addition every time MS patches, Parallels has not answered the question how they keep up. Yes MS could fix that, but..... Win 8 plus at best, and by then VDI is over. MS has no reason to move. Only real benefit that could get MS to care is horizontal server scaling as a feature of Windows for process isolation. Not sure MS is that smart, and if the MDOP team ever got hold of this, then for sure it's an eight year wait until they find a use case for it that nobody wants like Med-V. MS can afford to wait and see who wins VDI and just buy at that point if they need to. The real battle for VDI is management not hypervisor. That is a battle that is being fought in the datacenter, and this would just be a distraction for MS.

Once Calista comes out, I am sure it won't support ESX and VMWare will have to answer. Perhaps they should look at this and come up with an answer to TS since MS is not going to support license to enable DAAS until there is a lot more pressure that impact their bottom line. I always thought VMWare screwed up not buying Provision Networks to have an answer to TS/Citrix.

Of course Paul and others who have no Hypervisor would love to start a new battle ground because they have zero market share in TS and anemic investment from their mother ships to stay relevant. I don't buy it for one second from these tier 3 vendors.


BTW, Ringcube is the closet thing I have seen to Desktop containers. Has anybody checked them out for this use case? Would love to get your thoughts.


I understand what you mean... but the TS guys have been struggling with this for a long time.

I guess people thinking about implementing would need to look very carefully at their application portfolio and see what works, incl talking to vendors.

On the plus side 3 times the consolidation, and little OS sprawl, and containers with templates where u push the apps out from. Those are some nice features.

Here is an article on Parallels VDI sizing:


I hope ringcubes vspace is better then mojopac 2.0 engine.

As i couldnt get Acrobat 8 to install, or another app that uses localised SQL.

Eg Kernel mode issues....

That said mojo is meant to have a driver framework system... didnt work for me.

Anyway need to try again on vspace to be fair.


Can someone from Parallels come into this conversation... we would love your input.


Ever since I stumbled across the VIrtuozzo solution, I have been a fan of avoiding the VCED license.  

In our early roll-out most of our issues are not related to running server code but running x64 code.  The only application that did not work at this point is the Server Admin Tools from Microsoft (queue Citrix XenApp to correct the issue).

Just for scope we have deployed roughly 15 applications with Parallels.

Most issues he have had is around the broker because  the 64-bit toolset is not as mature as the 32-bit tools.



I was able to bend Mike Larkin's (founder and CTO of Ringcube) ear while at BriForum.  After many beers tossed back, I can definitely say it was a blast chatting with him and I've got a ton of respect for him as he's clearly a brilliant guy.  I had looked at the early Ringcube MojoPac stuff, but dismissed it for corp use primarily because of their user-mode level operating level that wouldn't provide any ability to ensure host integrity.  The new vDesk stuff operates a bit deeper and does allow things like VPNs and you could even deploy your own A/V stack, etc. but it requires some additional fiddling because the kernel I/O components aren't virtualized out of the box (but this can be enabled).  Anyway, I see a lot of promise for vDesk to complete in the client hypervisor space.  They have some of the delta technology like the other players, but what makes them unique is direct hardware access so there's no nonsense of emulated crappy graphics, etc.  Anyway, I'm not entirely sure that vDesk and Virtuozzo should be directly compared, because currently vDesk isn't targetted at multiple users on the same endpoint device.  They're more of a 1:1.  Could it be used in that model, I don't know but that's certainly something to look into.  However, once you have multi-user tenancy on the same OS (especially things like desktop OSs, I think you'll start to get the attention of the MS Licensing police pretty quickly.

Good discussion.



Appdetective, drivers must be whitelisted by Parallels for use in VZ Containers.  One this is done, drivers work fine on VZ.  This is done to ensure that the system is stable, since all containers share the underlying OS Kernel.

There are 4 primary reasons why companies look at Parallels Virtuozzo for VDI:

1. Very high scalability - similar to X64 Terminal Services

2.  Very low storage requirements, as all containers are based on based on the underlying OS (2003 X64) and contain only COW (Copy on Write) pointers to the file system & registry, thus they use ~ 50MB per container.

3.  Licensing - Since Parallels Virtuozzo is NOT A HYPERVISOR, DOES NOT VIRTUALIZE XP/VISTA... but rather installes a virtualization layer on top of Windows Server, it does not fall under the VECD Subscription Model.  It is licensed with a Windows Server Datacenter Model + TSCAL per device/user.  This is particularly advantageous to hosting providers under SPLA Agreements with Microsoft, who can't enroll in VECD under their agreement.  This is also advantageous to companies who have not chosen to take advantage of the benefits of Microsoft SA or EA agreements.

4.  Application delivery - Virtuozzo has a built in application templating mechanism that allows an application to be packaged sinimar to how it is done with an MSI packager. When deploying the apps, the applications are not copied to each container, but rather pointers to the application are written to the target containers, which uses virtually zero disk space per container, and happens in seconds.

The difference between containers and RDS Sessions are that each container has:

MAC Address

IP Address


Virtualized Registry

Virtualized File System

Complete Isolation from other containers

Users, regardless of permissions or rights, have absolutely no way to write to or alter the Virtuozzo Host, whereas in TS/RDS an administrator can destroy the system.

I was really hoping Microsoft would have purchased SWsoft, who merged with Parallels.  If this were to happen, Microsoft would have a stranglehold on the virtualization market, as this technology is off the chart in technical capability.


I also like Virtuozzo, very high density.

But i am also thinking that if you have a Windows datacenter lisense, could you provide W2k8 desktop to users without VECD lisense (ESX, Hyper-V). Only Windows CAL and TSE CAL needed?



if you remember this was the core of the presentation we gave at MVP Summit two years ago. Each of these suggestions was handily dismissed, although it did seem to somewhat influence them to increase session isolation in some smaller ways.

The specific feedback from MS at the time was that they looked at Parallels Virtuozzo Containers and concluded it was too much of a hack, that they could not incorporate these approaches into the mainstream OS itself.

My guess is that they come to understand and care about containers that they will right some version themselves in the next major Windows OS version, but as of last summit that did not seem to be on the roadmap.....


Parallels containers is an excellent case for VDI but it still does not takes care of the all the issues why you would move from TS to VDI;

1. Application Compatibility: Application which does not installs/work on Windows server platforms because fof platform issue are still not going to work with Parallels container

2. Driver level virtualization : Parallels containers can not virtualize kernel drivers. So if user installs a software which silently installs a kernel mode driver, it  may effect other containers. Though i have not tested it yet how it effects the system.

Parallels container does provide some unique features:

1. Move app templates across hosts dynamically

2. Assign apps, hardware resources in seconds without rebooting

3. create new containers in seconds without a need to worry about sysreps etc.

FYI: I work  for Propalms and we just released our new connection broker for Parallels containers.


I for one hope MS doesn't buy Parallels.  Why is it that you think development would accelerate under MS?  They didn't invent this technology (they didn't even design a standard hypervisor until recently) and they certainly wouldn't  continue development of the Linux version (or open source version; OpenVZ which is the basis for the commercial products).  You know, everything under one roof is almost never the best solution for a customer.  Here again is the 'embrace, extend, extinguish' strategy that saw MS confronted by the DOJ.  Parallels is doing just fine with the product and I personally doubt MS will consider a purchase.

Other items: 1) Parallels updates the OS from it's own 'update' servers...this is by design to ensure compatibility with Windows updates.  Parallels tests Windows updates prior to release...just ask. 2) Kernel drivers...duh.  It's a single virtualized kernel shared by the containers on a single physical host.  It would require another 'contained' kernel...and now your running multiple kernels and the licensing model goes out the window(s)  3) You cannot use Windows Server 2008 Data Center Edition with VMware View to escape the VECD license (see VMWare EULA).  4) TS inside of Parallels...I don't get the use case for that?  5) According to the agreement between Parallels & MS; MS officially supports it's applications in Parallels containers. I see this argument all time: it's a server OS, not a desktop OS, my 3rd party app will never be supported.  Yet, I rarely here that same argument with terminal services or citrix...and ya ain't using a desktop OS for that!


Would VMware be interested in them?


FWIW the new vDesk goes much deeper into the OS and supports kernel drivers, etc.  It's sort of like having two kernels running side by side.  So it does address many of the issues that existed in the original Mojopac product.  I'm waiting on getting the new code from Ringcube so I can kick the tires on the new 2.0 code.



Hi, Parallels here.  

We always enjoy seeing positive speculations about us; it’s very pleasing to be considered so desirable!  Hopefully it means we’re doing something right with our technology and it seems like some of the people reading this article think so.  A very big thank you to all those that posted in support of us.  

We want to take the opportunity to address some of the questions raised about Parallels Virtuozzo Containers... Brian summarized the approach and benefits of the product very well, as did Patrick Rouse, so I’ll try to avoid repeating too much of the info they provided.   Regarding the matter of running the server OS rather than, say, XP.  This is correct, but Windows Server 2003 and XP share a common kernel, which helps to ensure application compatibility between the two distributions, while end-users get the same “desktop” experience.  On the flip side, one of the advantages of running Windows Server rather than XP or Vista is that you don’t get hit with Microsoft’s additional VECD licensing fee, greatly reducing the total cost of ownership of Parallels VDI.  For anyone interested, you can learn more about VECD <a target="_new" href=">

If anyone wants more information on the solution, you can find it <a target="_new" href="> or send questions to me at  


You can layer PVC on ESX or Hyper-V.

You can layer Zen on ESX or Hyper-V.

Complexity is proportionate to layering virtual apps on virtual desktops with similar conciderations. Moving load from one layer to another to find the balance for your implimentation.

Both Zen and PVC were built upon GPL.

Princeton and Bell both released publications comparing the two but the last I've read are dated now... :(


@parallels. Win2k3 and XP are not the same kernel. They are versions 5.2 and 5.1 respectively so that is flawed argument. Win2K8 R2 and Win 7 are the same Kernel version number. Also a 'Desktop' experience is from a Desktop OS. Anything else is a compromise. Ok for many, but I can use TS for that compromise and MS is making TS better over time.

Unlike RingCube, Parallels forces you to share the same Kernel. RingCube virtualizes many of the Kernel components on a Desktop OS, and runs on top of a Hypervisor to support personalization.

If MS is buying this model and it goes against the Hypervisor model that so many are investing in, I don't buy that anybody is really going to do this. It also hurts MS license revenue, so there is NO way this is going to last long term.

You can implement VDI/Desktop virtualization on local storage, so that argument is BS also. Of course vendors like Quest want to scream this is great, because they don't have a Hypervisor. Who's actually implementing this at scale. How do Parallels sales compare to VMWare/Citrix? The market is just not there, it's a theory.

Finally, I don't buy Parallels can keep up with the rate of MS patches or assure me that there stuff won't break since MS does not test it. It's is in fact a hack. Similar concerns with RingCube, buy they have better technology.


@Shawn Bass


This thread is interesting and you guys are posting some insightful comments.

You two are both right that vDesk 2.x goes pretty deep and much deeper than anything ever built for consumers on MojoPac. The underlying WVEs (Workspace Virtualization Engines) are distinctly different in MojoPac 2.0 vs. vDesk 2.0 which is why we created this comparison chart:

We are trying our best to reach, explain, and show our software to as many of the desktop virtualization thought-leaders as possible.  So far the folks who have spent a good deal of time going "deeper" into our vDesk 2.x architecture are: Enterprise customers who have virtualization architects, Chris Wolf, Neil MacDonald, Brian Gammage, Leslie Fiering, Mark Margevicius, Adam Jaques, Harry Labana & team, and from what I'm learning from Mike Larkin - the list should probably now include Brian Madden and Shawn Bass (as a result of BriForum).

I can't really say whether or not our vDesk technology is better than Virtuozzo since we focus on different problems/platforms (server vs. desktop). But I can say that Parallels has always been a great partner to work with. And among our joint customers, I know folks in the banking / financial services sector absolutely swear by Virtuozzo for both their development and production servers. According to those joint customers, their "server performance and scalability is phenomenal" and better than what we can get with a type 1 hypervisor.  BTW, the workloads these customers use Virtuozzo for don't require live migration / v-motion just really high density.

Bottom-line: vDesk's heritage is and continues to be focused on the desktop OS. We do our best to build virtualization that's lightweight, low overhead, high performance, deep isolation, and highly mobile (PC, Drive, Network, VDI). Virtuozzo's heritage is on the server OS.  Both OS platforms are challenging to deliver really outstanding virtualization on.

If we ported to a server OS or they to a desktop OS, then I think a comparison would be valid but for now it's all pretty academic.  And yes, multiplexing a desktop OS for concurrent users is technically very possible with our technology and we're confident a certain segment of customers would love it but legally we'd have to get the "blessing" from MSFT (non-trivial) b/c licensing and the potential of negative financial impact on desktop OS sales could exist. Today, we enjoy having a very friendly relationship with MSFT, they help us, and we'd like to keep it that way. ;-)


@appdetective - your statement "RingCube virtualizes many of the Kernel components on a Desktop OS, and runs on top of a Hypervisor to support personalization." This is true and cool to see that you "get" our technology, sadly not everyone "gets it."

FYI, Larkin wrote a whitepaper comparing desktop virtualization architecture approaches - it's a pretty deep technical read and requires a lot of virtualization knowledge to fully appreciate but hopefully it sheds light on why we think WVEs are so useful when addressing desktops vs. servers.  Please feel free to pass it along if you get into debates or just need a pointer regarding architectural approaches:

Hope this is helpful.


Oh yeah, forgot to mention, in case some of you are curious the vDesk 2.0 exclusive product review was just published online:

Chris Wolf conducted the product review for Virtualization Review Magazine.