After VMworld, VMware still has to prove they "get" desktops. 13 questions about their desktop strategy they need to answer.

I wonder if maybe this is just a sign of the immaturity of this market overall.  I know Brian believes that VDI will hit mainstream next year, but I continue to believe that it is just too soon.  

The tools are not in place by any vendor to have a polished product, and everyone is learning as they go.Sure, there are some great peices that are supposed to appear soon, but some of these will be delayed or reduced in features.  We will probably also have to re-think some things based on realities learned by early adopters as well.

@Brian, thanks for including many of my questions. It's good to know to you get it!

@Tim, I agree with you, it is a sign of immaturity. However, I repeat what I have said many times.

People will not implement the pooled model at scale for a while. I agree with Tim here, it is not ready, does not scale etc. Citrix sucks at this also, unless you are a very simple implementation. Ask Citrix to show you a single customer that has implemented pooled successfully at scale that has any variance between users? The XD team put out crap quality products as well, we all know it, try installing the freaking thing, install a patch and see how that goes. Their value is HDX/ICA not that POS they call a broker. Perhaps the XD team can come forward and explain why they built such a POS infrastructure? I'd be questioning the judgement of the XD lead leadership team, idiots that don't get it either and focus on marketing not quality. So we should balance our view with what VMware can't do, who are just stupider, arrogant and I 100% agree with Brian not going for gold in this space.

We need the layered cake model to really work for this to be real. This is the money shot, but needs investment from the vendors to make pooled work. If not, then simply stick to TS for low cost environments, or wait for client virtualization to mature, which is also not that useful without layered type management.

People successfully implementing VDI today, are doing it for agility, and making a bet on the future. Those people are not implementing pooled, they are using private images, and yes it COSTS more, but they want pooled in the future if it can scale. If not, the benefits of layers still applies to static desktops that enable agility. To implement that simple model we need HDX Connect, when are the XD team going to wake and understand that POS middle tier that does not scale is going to take time to mature, change etc, so we need options to implement now!

Agreed - VDI isn't ready till 2011.  

Also, I think a datacenter guy isn't that bad a choice for  the next year when you think about the largest obstacles facing VDI right now: display protocol speediness and scalability (storage performance, storage de-dupe, user density, ..).

Once he get's these nuts and bolts issue solved, they can get a "desktop" guy and do more "desktoppy stuff".  :)

@Sean, the problem is that the myth is VDI in private mode needs $$$$ storage. Local disks works just fine and are cheap cheap cheap.

I see people saying that VDI is not mature... just wondering if we are not confusing "VMware's VDI solution is not mature", which I am not even sure I agree with, with the more generic "VDI is not mature". For example, looking at Citrix XenDesktop, I wonder how many people can say that this is not a well-thought over solution? Yes, it is piecing some separate products together, and yes, there is no single management console, but what did you expect? Isn't this also the case in the physical world?

AFAIK, there is no single-vendor A-Z solution for VDI (covering hypervisor, broker, OS management, user state management, user workspace management). You'll always end up combining different tools to get your own virtual desktop infrastructure, just like you would with any other SBC solution (combining TS, profile management tools & application management tools). For me, that doesn't seem related to the maturity of the solution but rather "expectation management".

I also disagree that the maturity of VDI is in any way related to the fact that a virtual desktop should perform exactly the same as a physical desktop. Probably very few people are surprised if you say that SBC's (as in: TS/Citrix XenApp) graphical performance is lower than that of a physical desktop, so why demand the same for a virtual desktop? Aren't we expecting too much when placing 50 virtual desktops on a CPU that is basically the same (or twice) the one I have in my laptop? IMHO, we're still talking about SBC and this will simply not always be a solution for all applications and all situations.

@appdetective  - agree on both accounts  - pooled is immature and VDI needs cheap IOPS that local disk provide.  

Amount of Storage is no longer  the primary issue  - IOPS and their cost is. The class of IOPS desktops require are simply unsustainable in price and performance from a T1/T2 SAN/NAS perspective.  Widestriping in SAN/NAS is not a cost effective approach for VDI.

Desktops have evolved  around assumptions  of low (1-2 ms) latency and easy availability of IOPS (think virtual memory, swapping,  Continuous NTFS file layout optimization).  The Shared, centralized SAN/NAS model cannot provide both of these at a price point  comparable  to a physical desktop.

We introduced vScaler  at vmworld 2009 - a technology specifically for addressing the needs for cheap and fast IOPS for Virtual Desktop and Server workloads and with the ability to use local disk as a part of the storage mix to bring parity in cost/performance between SAN/NAS and local disk.

The one BIG Issue in my mind that no one talks about (or have we just not realized it yet?) is that the the desktop IT guys have very little control and leverage over the infrastructure that runs their desktops when it comes to VDI.  With VDI, the desktop IT team simply own and control the content inside the VM - the VM itself and everything under it is locked away from the  desktop IT team.  This lack of control translates to an inability to dial-up/dial-down critical resources (memory/disk/IOPS) that affect the committed SLA and ultimately the end user experience. This can be frustrating for the desktop teams where they are asked to fix things they cant control. Its going to take some new enlightened organization and practices to be able to get around this one!

Chetan Venkatesh

Atlantis Computing

www.atlantiscomputing.com

www.twitter.com/chetan_

@Chetan, I agree with you 100% on the org stuff. However I have to give credit to Citrix for showing they get this issue community.citrix.com/.../Desktop+Virtualization+is+not+Server+Virtualization

As for the your solution, good to see you making progress. Watching keenly to see how you do vs. Unidesk. Great to see thought leadership out there.

I’ll throw out a couple of comments on the questions that are most relevant to my role at VMware, particularly ThinApp.   But first a question for you – Why 13 questions for VMware?  It seems like several of these are great questions, not for VMware, but for independent thinkers, consultants, and IT professionals to spur analysis and thoughtful design planning, and honestly common sense is all that’s necessary on some of these as well.  Anyway, my own take on some of the questions below :

Does VMware think that ThinApp can handle all apps today? Should they be virtualized with another product? (appvirt can’t do every app, just like TS can’t, but we(ThinApp) have the best out of box compatibility rate and if you cant do it with our product then it cant be done with another pure appvirt solution, more than likely you’ll need Ringcube or something from workspace virt) What is the official recommendation for people who have apps that aren’t ThinApp compatible? Should they all be installed natively? (answer – If the app isn’t virtualized then what other options are there besides native installation in desktop or TS OS?  The topic of ThinApp compatibility which this questions raises is the reality of whether the app can be virtualized depends most on the skill of the packager, in-depth knowledge of the applications, and possibly cooperation of the vendor; not the technology of ThinApp itself.  Just some additional commentary – often appvirt delivers immediate value in physical environments of existing desktops, so there are going to be native applications and that’s why appvirt isn’t meant to be an all or nothing approach most of the time.  There are going to be opportunities for a completely blank OS container more often in non-persistent virtual desktop environments and maybe in some physical desktop cases.  But for the most part Appvirt is a play to reduce operating cost of environments not the means to completely remove native applications.  You’ll see consensus amongst the appvirt vendors on this topic and if all the vendors agree then . . . 

Should View only be used where we have 100% ThinApp compat?’ (answer – that is such an all-or-nothing question that I doubt anyone reading this blog needs an answer to this, but obviously we sell View without ThinApp and clearly position ThinApp with View as a means to make the desktop more stateless and still maintain app settings persistence.  Certainly not every View desktop needs that capability and keep in mind that ThinApp alone doesn’t make the VM stateless, there is View Composer and the RTO technology that will further the use cases in this story as well.  

What’s VMware’s recommended technique for deploying ThinApp apps into VMs? Let’s say I really want to use View for everyone. How are apps delivered and managed? (answer – apps can be streamed or deployed, both offer pro’s and con’s, see our ThinApp Deployment Guide or the ThinApp Reference Architecture for a more comprehensive discussion on this but its not that difficult.  If you are using View Composer and recompose your destops monthly for patching then putting thinapps locally in the vm costs you nothing from an update perspective and has no impact on network and delivers the best performance.  If you’re application is updated very frequently then streaming is your best option from an administrative effort standpoint but do you have ample network throughput to ensure acceptable performance?  Streaming versus Deployed mode depends on the tools, processes, and environment, most often a hybrid approach makes the most sense.  So the answer here is RTM for details or put on your thinking cap and ask yourself the necessary questions to make these decisions.  There is nothing vendor specific about this process, some of the best material I’ve seen on this was from a Network World preso by an analyst at the last Citrix Synergy event.  

Does VMware believe that there is any use for server-hosted apps which could be delivered seamlessly into VMs? If so, what’s their recommended product for that? Citrix XenApp? Quest vWorkspace? Microsoft RemoteApp? (answer – why would you ask the vendor this question?  A customer can answer this perfectly by considering the time, effort, cost of supporting those other platforms.  And yes, VMware definitely acknowledges that there are a substantial number of TS/Citrix applications out there with a lot of infrastructure and people supporting them so if and when those apps migrate into local vm’s is a determination to be made by IT.  XenApp or RemoteApp, well, with vm’s just doing the display why would you pay $300 ccu to run ICA across your LAN?

Geez – what a treatise – so sorry

Now my response to IMO the best question on the list - Why does VMware think they can win a battle of the desktop against Microsoft when Microsoft owns the operating system? Isn't that reason enough that smart customers should sprint towards Microsoft+Citrix or Microsoft+Quest?  (answer – would just rather sit down at Giordano’s in Chicago for a couple of hours to discuss this.  I’ll be as concise as possible – MS does own the desktop OS (which is different than your Springsource question above) and that is why it is so ironic for them to offer a solution.  Does anyone else find it odd that MS sells MDOP to more efficiently manage the OS that was marketed as being more efficient to manage? As for Citrix, you already nailed the competitive pressure thing, I was working at Citrix about 6 years ago and was so jazzed when the CTO told me that we were going to ‘do it’ referring to an ICA desktop pass-thru thing.  So I was going to be able to sell a low cost, no infra solution to pass-thru access to desktop applications.  Sounded great, throw the cannibalize PS issue to the wind, 3months later, he was out the door, and three years later AFTER VMware introduced VDI, Citrix comes out with something very similar.  So my question is why you would expect MS or Citrix to offer a truly great solution for customers when they have so much to lose.

@appdetective –

I am on the leadership team for XenDesktop at Citrix.

I am sorry to hear that you had a bad experience with XenDesktop implementation. Based on all your posts, I have to assume that you understand the product and have experience in implementing Citrix’s products.

First of all, I want to acknowledge that we have heard the feedback regarding complexity in configuring advanced XenDesktop implementations that deliver single instance management with pooled desktops. However, we have also seen some of our customers successfully implement a ‘layered cake’ or single instance management with XenDesktop pooled desktops successfully. We have learnt a lot about where such advanced implementations succeed and where they may not. And, we are taking all this feedback to further enhance our product. We are looking at radically simplifying the management with XenDesktop delivered pooled desktops using single instance management in the near future.

Thank you for your candid comments about the product. We have no intent to over-market the capabilities of the product. We think that the product is capable of doing single instance management; however, I understand that with all the set of powerful capabilities customers have run into issues with complexity of setup and maintenance. I (along with my peers) would personally like to speak with you and better understand the issues that you ran into and see if they were similar to what we have heard so far. I can’t promise that I will be able to address them all during our conversations; however, I can definitely promise that we will use the input to quickly enhance our product AS QUICKLY AS WE CAN.

@Sumit, no need for us to speak. Just read the Citrix blogs and understand that in the real world at scale pooled desktops is not pragmatic for a long time. I can only implement persistent desktops for a long time, and for that I need the HDX Connect stuff that has had some discussion. If I want pooled it's easier for me to use XA, it's much simpler and a better product. For this class of use session isolation does not matter as much, plus grabbing the the inventory data to drive pooled is a *** at scale.

That said, this is not to say that I don't think you guys shouldn't invest in pooled and invent layers the right way for both XA and XD. Just understand pooled is one model, and I fully expect for large customers this to be the exception for a long time.

Pooled is about cost savings, so long term makes sense for the masses, today you have XA for that if you really need it. Today is you need session isolation with agility persistent is the only way, and putting a broker in the middle as the only way to use the product is a failure on Citrix's part to understand real world implementation.

Sumit, I hope you act AS QUICKY AS YOU CAN.

After being in an "In the deep" PoC phasis with VDI I'm not surprised every vendor wants' to be the chief when it comes to centralized desktops in the datacenter. Today's state of the art VDI solution is at my opinion:

- XenDesktop as Broker

- Provisioning Server for desktop image management and disk space saving

- VMware VI / vSphere for delivering the hypervisor platform

- Appsense Environment Manager with User Personalization for user profile management

If we look now on the cost side, even if we forget the microsoft licenses, there are a lot of vendors who want you to sell their complete solution. But in fact you have to pick out the best of each one. Of course, this is not unknown if we look back a few years regarding performant old school "Citrix Metaframe" environments.

But the point is: Which vendor will be able to provide a more or less "complete" solution for fullfilling all the requirements of a VDI infrastructure. At my point of view, there will be no one. It comes out to that what it was always, you have to carefully engineer your requirements and pick up a design which mostly combines the advantages of each product / vendor. At least I don't beleive that VMware is in the position to deliver a perfect solution beside their really not bad hypervisors / management tools.

VDI like most other technologies won't always be a packaged solution.  Just because Citrix did it for years with presentation server doesn't mean the same thing can be done for VDI.  They are two differnt technolgoies that solve the same problem.

However, the biggest problem with VDI is storage vs. the non-persitent pools of images.   The storage will cost a fortune and non-peristent pools have issues as well.  So what to do.

Do neither.  Use ESX as the infrastructure, put Atlantis Computing in to solve the storage problem (big banks are doing it) get a broker of choice, utilize virtualized applications (I prefer Symantec) and get a profile solution.

Not a widely complex solution but not that bad either.

Brian spoke of Atlantis Computing in prior posts and they do seem to have a nice product which reduces storage 10X to 20X and reduces IO on your SAN or NAS.

Just my two cents.