Microsoft hates Type 1 client hypervisors. (A guest post by AppDetective)

Great post!  hope this is just a first gurst post in a series of many. Seems like you've covered most of the bases, so I just want to highlight a couple of things;

Firstly I really see the need and value of a highly optimized and transparent type 2 hypervisor, Secondly I’m indignant to see that Microsoft is still the same fat, sorry “rich”  pc, client/server lock in company they’ve always been.

The other maneuvers never quite feel warm, not as a company strategy at least.

In that respect I wholeheartly believe that due to the power and position of the Microsoft giant, we need not one Goliath, but an army of them to force change. I believe this is good for the customer, good for the industry and, not least, good for Microsoft.

Good to see you as a new poster at BrianMadden.com, appdetective!

Are you aware of Windows Server Embedded? It supports virtualization, so Microsoft is theoretically lined-up for a near-type 1 ("type 1,5") hypervisor.

Basically it's up to the OEM's out there to bundle it with their existing hardware. Server Embedded probably lacks support for mobile users though, and is probably also to expensive. But what keeps MS from releasing a mobile edition?

Actually - I think that vPro isn't a hard requirement for XenClient or CVP but something that will allow deeper management and security options  for the post 1.0 releases (so is strongly recommended by the vendors).

Are you kidding me? T1 client hypervisor has the potential of killing Microsoft. Microsoft under Ballmer has not responded in meaningful way to any emerging technology/threat, they are reactive to everything that happens around them. The only thing that keeps their market cap so high is those corporate PCs running Slowdows. Enter T1 hypervisor that can play music, control DVD burning and do everything else with hardware that the OS does now and all of a sudden the desktop management reality changes. Microsoft will not have a T1 hypervisor because there is NO WAY they can do it right and own that market (that will go to Vmware) so by introducing their own T1 would just validate the market need. They will do what they do best: stall the market by influencing OEMs so they can pull one more Windows refresh cycle.

Insightful comments.  I agree that MS will do everything to protect their control over the desktop, however they may not have much of a choice but to offer some form of client hypervisor, especially if they want to support server-based and client-based virtualization in a seamless way.  I doubt MS bought Sentillion for their vThere product as we have almost never run into vThere anywhere, even at our healthcare customers.

At MokaFive we purposely architected our system to be independent of the VMM layer and to work well in a Type 2 as well as a Type 1 world.  We saw that the VMM layer was increasingly becoming commoditized and so we focused on the management aspect.  Today we ship with a Type 2 solution that is robust and enterprise-ready.  We are experimenting with Type 1 solutions but as of today Type 1 technology is nowhere close to something that we would want to support in a real deployment.  Maybe in 2010 the technology will start to get there, but it is still quite new and has a lot of dependencies on hardware manufacturers and device drivers.  The HCL for the early client hypervisors will be very, very short and that's why the dependency on Intel's vPro and hardware refresh - Intel is supplying the drivers.  We have plans to officially support a Type 1 solution once the technology matures a bit more and we feel it is something we can support.  But I wouldn't expect major deployments in 2010 or even 2011.

Going forward, I think there will be a place for both Type 1 (e.g. corporate-owned machine) and Type 2 (e.g. individual's machine or BYOPC), and any management solution worth its salt will need to be able to manage both types of clients.  The distinction between the two is also fuzzy because you can do hardware passthrough on Type 2, and with more functionality and management being built in the Type 1 base it is beginning to look like a full-fledged OS.  VMware's CVP is a Type 2 architecture with a Linux host OS and VMware Player on top.  

If Microsoft is smart they will do a client hypervisor so they can control that layer too.  They definitely have the technology, resources, and clout to do it.  They also have all the device drivers, which has been an Achilles heel of client hypervisors thus far.

John Whaley

MokaFive CTO

Needless to say, at Virtual Computer, we are a bit more bullish on the maturity of type 1 client hypervisors and the new use cases and management wins they provide. There is a lot to discuss here, but for what I expect will be the first of several comments to this post, I am going to start with hardware compatibility.

Hardware compatibility is certainly one of the bigger challenges with client hypervisors, but is actually not as hard as people perceive it to be.  The focus on vPro is really a bit of a red herring.  It really comes down to whether a PC has Intel VT-x or VT-d (or AMD-V as the case may be).  If a vendor is doing things like pass-through graphics, it is easier and more secure to do this with VT-d.  Since VT-d is generally bundled with vPro, that is where the association generally come from.  Done “right,” as we think we have, all you need is VT-x, which has been on chips for several years, is generally necessary for Type-2 anyways, and doesn’t tie a Type-1 hypervisor rollout to a hardware refresh.  I do all of my product demos on a Dell D630 that is several years old and does not have vPro/VT-d.  I use it because it is a familiar “workhorse” platform for many organizations we talk to.

When it comes to drivers, it’s certainly not trivial, but it’s not splitting the atom either.  If you take USB, which we pipe through to Windows, out of the equation, there really aren’t that many drivers to worry about to cover the “big 3” corporate PC vendors and many others as well.  On graphics, we have had Intel drivers since day one, Nvidia since last summer, and have ATI support close in our sights.  On wireless, we have Intel, Broadcom (and Dell’s OEM version thereof), and Atheros covered.  And the list goes on…  In any given driver category, there are really only two or three that matter, most of which have existing Linux drivers available that can be leveraged in a Xen-based client hypervisor architecture.

Quite honestly, the biggest challenge with respect to hardware compatibility is the “last mile” of smaller details.  For example, different platforms communicate ACPI events around things like laptop lid closure, etc. in different ways.  Slightly annoying and inconvenient for software vendors like us?  Yes.  Rocket science to deal with?  No.  We analyze how events are handled on a particular PC, make some basic changes if we need to, and move on.  Our biggest challenge as a small company is getting our hands on a wide enough breadth of platforms to analyze the differences.  Fortunately, there are a number of forward thinking PC OEMs out there who proactively ship us gear to test.  Dell in particular has been outstanding in this area, and in fact my D630 demo system actually has a “Property of Dell” sticker on it.  

That’s it for now.  I am going to resist the urge to take the bait on the “limited traction” and “lack of mature management” barbs, though I reserve the right to change my mind later.  :)

Doug Lane

Virtual Computer, Inc.

We know the DeviceVM folks well and have a lot of respect for them.  As a small company, they have done an amazing job of getting traction with the OEMs and actually bringing stuff to market through that path.

We certainly spend a fair bit of time engaging with the OEMs ourselves, and these efforts are really starting to bear fruit.  However, in my view the best route to getting traction with the OEMs is to go out and engage with enterprises and let the OEMs hear from their customers that client-side virtualization matters and is a major buying criteria.

Isn’t this kind of how it unfolded on the server virtualization side?  It wasn’t like VMware built their server virtualization product and said, “let’s wait for the server manufacturers and Microsoft to bless this and give us a path to market.”  Never would have happened.  They went after it with a good product, made themselves relevant in the enterprise, and now when you do go an order a server they are a line item option.  You won't find anyone further from VMware fanboy than me, but I respect that, and we intend to do exactly the same thing on the client.

I agree that eventually client hypervisors will ship from the factory on the PC, but I don’t agree that it isn’t a real market delivering real value to enterprise customers until that happens.

I'm gonna sit back and eat some popcorn.  This is going to get good :D

I take Doug’s point on engineering in hardware compatibility for some specific machines but for mainstream deployment I think it will require a hardware replacement cycle before client hypervisors become common. Here is my thinking:

Organizations will want support from their hardware vendors for machines running client hypervisors. That is doable for new machines but it is not in vendors’ interests to test and certify the back catalog so that customers can avoid buying a new machine (ignoring any driver issues).

Even if all existing machines were magically supported people would still be unlikely to roll out client hypervisors. Once machines are out in the field and working then it would be a huge operation to re-engineer the platform and then bring all those machines back through IT. Major changes like this are far easier to do during migrations and refreshes.

Hence, much as we might want client hypervisors on every machine right now (given all the other technology was in place) it is not going to happen – hardware vendors won’t support it and organizations would not roll it out.

Martin Ingram (AppSense)

@Martin

So organizations will wait to capture the benefits of client virtualization until the hardware vendors have announced full and unequivocal support for hypervisors the same way it all unfolded with server virtualization?  Oh wait, that’s not how it unfolded with server virtualization, is it?

VMware, XenSource/Citrix, Virtual Iron, etc. were all out there being adopted by some very large organizations for years before any official support came from the server OEMs.  The business benefits and ROI were simply too great to wait for.  The vendors just engaged directly with customers (and the OEMs’ own field sales teams!), OEM sales team were empowered to win deals / stand behind the solutions it took to win them, and enterprises pressed forward.  Eventually, standard offerings and support policies caught up, but only after server virtualization was a proven winner with major accounts.

We’ve seen this movie before.  Time to film the sequel.

That said, I don’t take the time, cost, and effort of reimaging PCs outside of a refresh lightly.  I do think that a significant number of customers will move conservatively over the course of a refresh cycle, and at Virtual Computer we have certainly done our business planning assuming that pace.  However, it’s worth noting that a lot of what has made PC reimaging painful and time consuming in the past can actually be avoided with client-side virtualization.  Beyond the fundamentals of no longer needing to worry about drivers in a master image, greater application deployment flexibility, etc., there are a myriad of mature technologies for things like P2V conversion and profile migration (which I suspect you probably know 100 times more about than me…) that can accelerate enterprise adoption ahead of the hardware refresh cycle.

@appdetective

The market factors, barriers, and likely-to-be-missed opportunities you describe are all very real, but that’s why god invented startups.  Someone has to stir the drink and just go for it in order for anything to happen.  And if startups didn't succeed versus larger, more established rivals at least part of the time, people would stop funding them.

I have seen enough positive reaction from customers and OEMs to feel pretty good about 2010 from my seat.  Time will tell I suppose.  Either way, our foot isn’t coming off the gas pedal and we have plenty of VC dollars in the tank (note to self: send X-mas card to Citrix…), so we’ll see how it all shakes out next year.

@appdetective

Fully agree that we don’t succeed until we have a Windows experience on the PC that requires no functionality or performance sacrifices for the end-user.  I don’t claim we are fully there, but we are closer than you think.  If we take your multi-monitor example, we have had mirroring for quite a while, and the latest code we are running internally (Q1 availability) has fully virtualized graphics with multi-monitor/extended desktop, all controlled right from within Windows without the need for a GPU-specific driver.

There may always be differences, but there will also be user experience attributes that are better on a client hypervisor.  As one example, I can run Windows 7 and Windows XP side-by-side on NxTop Engine, and if I close the lid, the VMs and the hypervisor suspend in a few seconds.  If I open the lid, the hypervisor and both VMs resume in a few seconds.  If I was playing a video when it suspended, the video picks up right where it left off.  I would contend that this is better than a native Windows experience, and we were doing this demo live at VMworld in August using our then-current NxTop 1.2 software.  (Very gratifying to hear VMware talk at the same show about how CVP would be able to run two VMs at once….someday.)

I also don’t agree that we are necessarily competing with System Center / MDOP.  There’s tons of stuff System Center does that we don’t.  There’s a ton of stuff NxTop Center can do that System Center can’t.  Put them together, and it’s like peanut butter and chocolate. NxTop Center runs exclusively on Server 2008, uses Hyper-V exclusively to create VMs and run them for administrative/patching purposes (even though we use Xen on the client side), and every single function in NxTop Center is API addressable in addition to being accessible through our UI.  We haven't completed the integration yet, but pretty easy stuff compared to what we are working on now.

I know you are going to hate this, since you obviously don’t like System Center, but in theory the combination of System Center and NxTop Center could actually give you a single pane of glass for native/non-virtualized desktops, VDI (assuming RDS matures), and NxTop client hypervisor based PCs (along with all the stuff it does on the server side).  Is VMware View going support VDI, offline virtual desktops, AND native PC patching any time soon?

If eventually, Microsoft adds a type-1 hypervisor and fills the System Center gaps we cover, so be it.  We’ll be on to the next thing.  Citrix has made a pretty good living at this in the Terminal Services arena.

I am so not getting any work done for the rest of the week, am I?

@appdetective

While it wouldn’t shock me if someone from MSFT trying to sell you something said something disparaging about Type-1, I don’t know that this stuff is even enough on the radar screen at the mother ship at this point to evoke hatred.  We can get to hundreds of millions in revenue before it makes sense for someone in Redmond to stop thinking about eating VMware lunch on server virtualization or driving Windows 7 upgrades for more than 5 minutes.  How long did they let VMware run for before they jumped in on server virtualization in a meaningful way?

Now imagine a company trying to do something very similar on desktops, but instead of thumbing their nose at MSFT, they embraced MSFT at every turn, including:

- Giving people a reason to upgrade from Server 2003 to Server 2008.

- Evangelizing/demonstrating Hyper-V with extreme loyalty, even when we are at VMworld or marching into accounts where VMware is entrenched.

- Driving Software Assurance purchases by creating new multi-OS use cases.

- Laying out an integration vision for System Center; not trying to replace it or talk anyone out of buying it.

What’s the urgency for them to jump in or try to squash us? Yeah, maybe someone decides to use us for Windows 7 / XP app compatibility instead of MED-V.  Who cares?  As long as we are doing all of the above, helping people get to Windows 7, and being any level of an annoyance to VMware on the desktop front, what’s to hate?

@appdetective, I would not be so excited about startups. I have worked for two before. At a startup the VC's *** slap everyone and ride them till they're dead. No one has any real power except the VC's. Virtual Computer is an early stage startup - no one knows if they will pull it through '10.

An advisor of ours pointed me to this very interesting thread. We certainly appreciate all the mentions with regard to DeviceVM.

It's clear from the discussions that proliferation of client hypervisors on non-Windows (the de facto case for Type 1) requires a solution to the issue of broad platform support.

Broad platform support has been an interesting by-product of our business of adding an instant-on Internet appliance function to as many PCs as we can. We work with OEMs directly; each PC that we pre-bundle with requires a Linux stack that has fully functional WiFi, LAN, SATA, graphics, audio in/out, webcam, USB peripherals, touchpad (sometimes with gestures), 3G, etc. We end up partnering with the various hardware device vendors to jointly ensure the Linux driver can work properly in time for the OEM's time-to-market.

There's a ways to go though. Native graphics drivers in Linux which would be important for client virtualization aren't always achievable due to time-to-market. Touchpad operation on par with Windows still needs driver improvements from device vendors. Many drivers don't move at the pace of kernel.org releases. Nonetheless, we'll end this year with tens of millions of PCs shipping with our instant-on stack (mostly consumer PCs). Hopefully next year we'll have that number alone for commercial PCs. Many of these will ship with an update agent built-in.

Which might make an interesting conversation with client hypervisor partners who're looking to get ahead of the enterprises' hardware refresh cycle and tap into a bigger HCL. (We don't have a hypervisor offering and intend to leave that to you experts.)

Phil / DeviceVM

The bottom line for me was finding a way to protect myself from users, viruses, spyware, and the windows death march to slowdom.  I tried switching users to Linux but it is still too hard for the average users.  I eyed perpetual live cds.  How could a user screw that up?  I looked at RingCube but I didn't like being tied to windows.  I decided to user Virtual Computers Nxtop.  The timing was perfect.  I was at the point where my newest laptop was at least 3 years old. The servers are older than that.  So I purchased laptops that were compatible with NxTop.  I'm bringing them online at about 2 per month plus a desktop here and there.

NxTop has saved me time pretty much out of the box.  One of my users got the Anti-virus 2009 virus a week after getting the NxTop.  I told her to restart the computer and the virus was gone.  Today I pushed out updated programs (anti-virus, backup program and windows update).  I did it all from the central management.  Updated 1 machine and pushed it out to everyone.

So, Type 1 or Type 2, I don't care.  NxTop's features and capabilities are good enough right now.  If they get better in the future because OEM Type 1 hypervisors, great.   Supporting 50 windows users with a single IT guy has become easier.  Bring on more users.....  Well maybe not.

Oh, and I definitely don't care what Microsoft does or doesn't like.  They lost me with Vista.