Prediction: 90% of the future "VDI" will be client-based

I'm not so sure.

One of the main reasons my company is pushing for SBC and VDI is as a partner to data centralisation - pulling in data from branch offices and roaming laptops so that it never leaves the datacenter.  Inherently this means moving the applications to the same place to maintain performance.

Client hypervisors may well be the answer for the minority portion of our users who need a laptop an can't guarantee connectivity back to the datacenter, but it's certainly not something I want to see being the de facto standard as it encourages data sprawl to maintain application performance.

Intersting prediction Brian.  But I have to ask; does anyone have accurate information on the number of 'simpler workers' worldwide?  I would tend to believe this workforce far outnumbers the 'power workers' who might require client-based VDI.

In that same vein of thought, I believe your wrong about 'everyone who really needs VDI already has it'.  I remember the transition from fat clients to Terminal Services.  From my perspective at the time, it took quite a few years for TS/Citrix to get where it is today.  I don't believe your concept applied to that transition, and equally so the transition to VDI.

I tend to agree with PhilG; client-based VDI will provide mobility to 'power users'.  And while mobile computing devices may have eclipsed non-mobile devices (I would actually like to see numbers on this in the worldwide entrpise), it seems that remote connectivity is (becoming) nearly ubiquitous.  Combine that with the push to centralize all data, and I'm just not sure if the need for client-base VDI is so strong in the long term.

I really like the idea of local VM syncing to a data center image, but I like the idea of centralizing all my data while providing remote connectivity, even more.  If (and that's a big IF) remote connectivity developes to be robust enough over the next decade, the business laptop may devolve into nothing more than a simple access device to the internal/external corporate compute cloud.

Then again, I could be completely wrong...

Brian - my informal poll of a dozen odd Financial Institutions says that 20-25% will be client virtualized while about 55-60% will be server hosted desktops by 2013.  while each sector probably will see a different adoption rate, Finance b'cos of security and compliance needs is tending to drive centralization beyond just the image and life-cycle to include execution.  

2 Questions I've been  asking myself  -

1. If a user could not distinguish between a VDI and client hosted desktop because of parity in performance, experience etc - what would IT want to give him/her?

2. if VMs are pure execution containers, and all the content (OS/Apps/user data) are centralized (or  accessible from everywhere) - where would you run the VM and why?

The answer in both cases to me seems obvious and  is the Data center. Here's why -

1. The Rack is the new computer.  It is now the NEW fundamental element of compute. (Cisco UCS is a bright indicator of this future )

2. 10G Ethernet is the new Bus connecting Compute, memory and primary storage

3. The Hyervisor is the kernel and VMs are simply threads/processes to be scheduled across the rack (or racks).

4. Memory Page Sharing, Primary Storage Deduplication, CDP, data mirroring, DR are new primary subsystems of the new Datacenter OS. Many  of these are astronomical in cost/complexity to do on the PC.

The PC/Client Type1/Client Type2 simply cannot compete with the awesome power, cost effectiveness, scale and efficiency of this new Compute model  (and elegance I might add).  

The PC Era is Dead - we need closure with this era, its architecture - and all the attempts to frankenstein it with Type 1/2 virtualization.   Lets have a  moment of silence for the PC era please.  Thank you.

Now we can move on and do some of the really innovative things that Virtualization in the Data center allows us to do for Windows, Desktops, Users and mankind.  

Best

Chetan Venkatesh

CTO & Founder

Atlantis Computing

--

PS - these are my views and not my employers.

I'm just not seeing Brian's vision of the future.  That "smart replication" syncing is going to take some time over remote connections.  Sure the "upload" side might be able to minmize that if the sync was continous.  But the "download" side (syncing up to the previous session's state) will delay a user from getting down to business.  And that'll invoke trouble, AKA support calls & resistance to adopting the technology.

I can see where a remote user may install a fairly large app.  And, sure, the delta gets "smart replicated" to the central host.  Now what happens when he uses another client?  All that delta data has to be downloaded before the user can use the app.

So I think SBC-based VDI will be the direction for quite some time.  No one has really demonstrated a solid use-case for Type-1 clients outside the area of the "power user".  We, as IT guys, often need to be reminded that technology is there to support business users who perform tasks.  There are a whole lot more of those "task workers" than there are "power users".

I completely agree with Chetan. I believe the future is "Cloud" not even enterprise data centers.

Desktops will be delivered as a cloud service [public/private] just like any other enterprise apps.

Type 2 and Type 1 will be special cases for power users.

Brian, we've been seeing this trend ourselves from our customers and leads.  We've been seeing more and more enterprises come to us for a client-side solution after determining server-side VDI doesn't make sense for a large swath of their employees.

The primary driver is cost.  VDI just doesn't make sense from a cost perspective.  Datacenter space and power is expensive.  Servers are expensive.  SAN storage is expensive.  I had a discussion recently the architect of a big manufacturer, and he had an interesting perspective: "if a laptop today costs $400 and comes with over 100GB of diskspace, does it make practical sense for me to move his data onto an expensive SAN?"  For him, the minimal cost of introducing a client-based virtualization solution made the decision straightforward - he could get the hardware independence of virtualization at much lower cost and risk than VDI.

Furthermore, server-based VDI flies in the face of one of the largest undeniable trends in computing: the progression from desktops to laptops.  Laptop sales have already overtaken desktop sales in 2008 and by 2011 there will be 2x as many laptops sold as desktops, according to Intel.  Thus, the traditional desktop management problem is quickly becoming a laptop management problem.  And laptops are often taken offline.

There are definite use cases for centralizing desktop execution in the data center, for example for high security situations where encryption is not good enough, and computationally-intensive situations where the computation needs to be near the (large) data set.  And of course, once you port your applications to run in the cloud you can get a huge benefit from running those apps/services in the cloud.  But for most use cases, I think executing your desktop locally makes more sense than executing your desktop in the cloud, especially from a cost perspective.

Products like TS and XenApp get you 90% of the benefit of server-based computing for a much lower cost.  I remember you coined "Madden's Paradox" a while back to describe this phenomenon :-).  The length in which the VDI vendors are willing to go to make the economics made sense is quite a stretch.  125 concurrent users per server?  So you are going from users who are used to 1-2 cores and taking them to 0.064 cores.  There is no magic that software can do to change that fact; the new Xeon processors are fast, but they are not that fast.  And if you have task-worker types who can get away with 1/30th of the processing power, I would argue they would be better served with a cheap CPU at the endpoint like an Atom (which by the way are *still cheaper* than thin client devices).

Incidentally, with respect to MokaFive, having an underlying OS is a BIG benefit for a key use case: BYOPC.  We’ve talked with many companies that are asking workers and contractors to bring their own machines.  Workers and contractors will not rootkit their machine for the company, and the company doesn't want to manage it anyway.  Surprisingly (or perhaps not), many employees are now using Macs, which makes them very happy.  And the IT folks too have been very happy with the performance and security of MokaFive.  We plan to support client hypervisors once they mature a bit more, so you can use the same image for BYOPC as corporate-owned and managed endpoints.

@john waley - thanks for reminding me of Brian's Paradox – Brian made an astute observation but  in the light of new and near term developments to come- not valid.  Its time to resolve the paradox.

TS gets its user density from partitioning the OS and creating containers to run user session specific processes, threads and IO.  TS's superior density compared to VDI really comes from the fact that there is ONE OS instance and multi-tenanted user sessions.  XA and clever technologies take this  further by single instancing some of the user session components themselves (sharing memory, dlls etc).

VDI of course does not benefit from this because each user session comes with its baggage of a dedicated OS and application payloads. Up until now.

New developments allow single instancing VM memory contents across a hypervisor effectively allowing just one copy of all the common objects across the entire memory space of a hypervisor. This means that  while there are 60-90 unique  desktop VMs running, from the hypervisor’s perspective all the common memory pages – kernel space and user space are de-duplicated and shared.  Yes that means there is only ONE instance of OS, working set, dlls, apps etc in memory. So theoretically if one shared image of  OS + Apps is booted to run 60 VMs, this optimization reduces the memory footprint of all those desktop instances to that of a single vm.

So with ONE nifty optimization you can now bring densities of VMs per core on VDI inline with users/core on TS.

VMware has already implemented some early forms of this with memory page sharing but future version of vSphere will advance this.  Redhat already benefits from having an excellent implementation of this through KSM (Kernel Shared Memory) in Linux 2.6.31 (KSM came from Qumranet and KVM). KSM was already available on Linux in Apr ’09 but since VDI is not usually on KVM this has not been proven yet. Going forward once Redhat releases the full VDI solution on KVM, this would be an enormous differentiator for them.  XenServer will inherit all of this  given the Open Source heritage of these projects and that means that XD gets these benefits on XenServer.

Now if VDI user density rivals TS – then what benefits does TS have other than cheaper end user licensing (and simpler/better understood best practices around install/manage) – not many technical ones.

That resolves the Brian Paradox

Now lets look at the following important constraints in VDI -

Datacenter space and power is expensive. Agreed but  PCs are no cheaper – just harder to consolidate and rollup the power and space costs to provide an accurate number. The thin client vendors like Pano and Wyse love to show how PCs are waste of electricity and real estate  and their numbers are defensible.

Servers are expensive – true – But in the same way as Boeing 747s are more expensive than GM Malibus. But 747s are long haul as is VDI – thousands of users and multi-tenanted.  Average CPU utilization of a PC is low – 0.064 core is about right if you average the utilization over a users day.  When a VM needs more,  hypervisors are smart enough to parcel out more cycles to handle bursts. This is known science and admins everywhere setup resource pools and reservations to deal with this.  When you have 60 Desktops hosted on a server that costs $10,000 – that’s $167 bucks a user.  That’s pretty cost efficient and inline with desktop costs (if not lower)

SAN storage is expensive – yes, but why use SAN when you can use aggregate local disk storage on servers.  Technology to allow VDI to combine local storage (for image) and SAN/NAS (for user data) is already here today – Atlantis’s Variable Volume Control (VVC) feature, Synchron, and in a more limited way - Lefthand vSAN, SAN Melody etc.  Atlantis has many  customers who are on-boarding  2k-5k desktops on VDI on local storage using VVC. They end up using about 1 TB of SAN for 1000 users.

Does it make sense for customers to do VDI the old way with every user having a full image footprint on SAN – NO (I agree with you!). But  Does it make sense for customers to use 1-1.5TB of SAN to onboard 1000 users – sure that’s damn cheap and you get all the reliability, DR and backup goodness of SAN. Those features  on distributed PCs (with or without Virtualization) are unreliable, expensive and complex to implement/administer.

Enterprise customers have tens of thousands of business critical LOB apps that cannot/will not be re-written as Web apps for the cloud in the next 5 – 8 years.  The vast majority of such in-house applications are not going to transform to a web stack because they work well enough and there’s no one to fix something that’s not broken. The old adage – the enemy of better is good enough holds true especially in risk sensitive IT departments.

VDI is going mainstream in the 2010-2012 time frame because for the majority of enterprise desktops (50%-60% as my customers tell me), VDI affords a great model of provisioning, self service, lifecycle management and break/fix.  The Client Hypervisors will benefit use cases on the boundaries of VDI where users need offline access because they work in remote locations where network, latency and b/width are constraints.

There is no reason why BYOPC use-cases don’t benefit from VDI – they might as well launch a VDI session when they need a desktop (unless of course they are the boundary users talked of previously).

So going back to my previous comment –

1. If a user could not distinguish between a VDI and client hosted desktop because of parity in performance, experience etc - what would IT want to give him/her?

2. if VMs are pure execution containers, and all the content (OS/Apps/user data) are centralized (or  accessible from everywhere) - where would you run the VM and why?

And one more to ponder

3. If there is cost parity between a physical desktop and a hosted one – which one would IT prefer to give to its users?

I think the balance tilts heavily in favor of  the Datacenter and Hosted Virtual Desktops.

Best

Chetan Venkatesh

CTO

Atlantis Computing

twitter: @chetan_

chetan@atlantiscomputing.com

Holy Book of Chetan, I think I might have to copy and paste that into word just to read. I fear that reading in such a narrow pane might also narrow my mind.

Sup Chetan!  Loved this bit "The PC Era is Dead "

I will add... Distributed Computed Failed!! All the mainframe user wanted was a mouse, we gave them Windows.....

@shanetech  howdy mark!  That was a long post now that you pointed it out. Hopefully worth reading all the way thru.

Distributed Compute Failed!! All the mainframe user wanted was a mouse, we gave them Windows..... <- Amen !

We need to petition  Brian and Gabe to make the comments pane wider.  

90% is a high percentage but since we are starting close to 100% for the desktop already, I definitely agree with Brian M that client-based computing will dominate compared to server-based for the desktop.

And I also agree with John Whaley that cost is the main driver on why there will be more client-centric virtual desktops versus server-centric for at least the next 5 years. The economics of VDI at very large scale just don't add up and most customers who have experience rolling out VDI solutions from VMware and Citrix know this too.

However, to think client-side virtualization will only come in the form of type 1 and type 2 hypervisors is having too narrow a view.

The problem with type 2 is that it has had more than 10 years of development yet still can't find a mainstream market use case for business users.  Don't get me wrong it's absolutely the best way to run Windows on a Mac OS X system AND great for a sandbox environment for QA testing and demos. But the large size, slow performance and licensing complexity that the type 2 model has makes it less than ideal for a typical business user and their respective Enterprise IT organization.

Type 1 client hypervisor is the "PKI" of desktop virtualization. Academics and theorists love what it represents on paper while pragmatics and operators scratch their head and wonder if it will ever be viable and pervasive in a timeframe that is meaningful.

Don't get me wrong, type 1 client hypervisors will have a impact on desktop virtualization. But just as PKI has an impact today on IT security.  The problem with PKI was that it became viable and pervasive about 10 years after all the hype of its introduction died down.

I fear that type 1 client hypervisors share the same fate as PKI. Great on paper but going to take a much longer time than even the most conservative estimates predict.

My prediction, type 1 client hypervisors will be viable and pervasive when Microsoft ships it's next major desktop operating system (i.e. Windows 8) until then it will have tons of promise on paper but come up very short in the real world.

There are other ways to deliver client-side desktop virtualization that's far more pragmatic than type 1 and far better performing than type 2. And there are several vendors who are innovating in that space.

These are exciting times in desktop virtualization with no clear leaders and the future is being written on a daily basis.

Great post and great vision. I couldn’t agree more with the assessment that the vast majority of “VDI” will be client-based, simply because (server-based) VDI, while providing many benefits of centralization, has two fundamental drawbacks:  1.  It doesn’t work for mobile/laptop/remote knowledge workers (majority of enterprise users in 2010), who need to be able to work offline as well as leverage the endpoints’ rich user experience and native performance that does not depend on the latency of the network or the load of the servers for every keyboard stroke that they make; 2. it requires massive and costly infrastructure build-out in the data-center,  duplicating the computing resources that are already available in client’s powerful processors, which under VDI become hugely underutilized by turning the powerful laptop into dumb terminal.

However, moving to a simple client-based solution impacts some of the key benefits of centralization such as compliance, continuity, protection and ease of manageability. Thus, the key challenge is how to approximate the benefits of VDI without sacrificing user-experience. Specifically, how to maintain the *full state* of the desktop in the data-center for management, protection, and support purposes, while allowing users to execute their workloads on the endpoints to leverage their compute power and provide adequate user experience. This is still a challenge that most vendors have not met yet, and is the key to bridging the gap between centralization and user-experience. The Wanova architecture and product aim at fullfilling this vision.  

Issy@wanova.com

As far as the debate over which if any hypervisor will rule -- both type-2 and type-1 hypervisors have their needs, and their pros and cons. In addition, many customers are not nearly ready or not interested to deploy either of those on their massive existing client base (e.g., lack of VT extensions in their clients) . Hence, an effective DV solution should support both types of client hypervisors, but not require any of them to manage the endpoint, i.e., it should be able to manage the bare-metal OS as well.  

Dear All,

Here is what I'm seing today : The Desktop reality is complex and Enterprises really need a solution that is able to handle this complexity.

Customers are really seeking for "The Right Desktop for the right population" and with the best TCO optimization possible.

That's why for now, on every big projects, we have task workers (50 to 80% of the total users) that will be provided a centralized XenApp Desktop (or published applications) environment and the rest (mainly about 20%) that'll use a XenDesktop Hosted Desktop (VM, Blade ...) or could switch to a XenClient one.

Of course Type I hypervisor for clients can add offline functionnalities but the extra costs will be accepted only for VIPs and roming users because they need such features to work, BYOC is not that spread and in my opinion, tasks workers won't be provided with client based VDI because of costs and management constraints.

Completely Agree..

Datacenter VDI is mostly (costly) Hyper as we encounter in our beloved love. Anyone remember ASP ?? (not the iis's asp..:) . But you have to be older than 20 year to remember it.

Virtualizing desktop like server is no-sens. Hyping VDI as Server's virtualisation is just another "build the next need to make $$$".

VDI is great for some specifique situations. Terminal Service do a lot of VDI's work better... but if you haven't TS in your portfolio (VmWare..) you have to make peple believe it's crap and that your product is great..

well... if today, there isn't 90% of the desktop which are virtualized (like gartner or >put your VDI's provider trademark> would like).. please don't think it's because 90% of the population is dumb...

it's the same thing as with.. say, the Iphone's success, Naruto's success... but of course just the opposite effect. if 90% of the people don't buy your hype/product, there is probably some reasons.. :-)

Interesting time anyway.  Best Regards..

about security and connecting to your session in the same state from anywhere, and the added benefit of others not screwing you as is the case with TS. It also allows a lot more personalization. Client Hypervisors don’t offer this unless somebody can virtualize the entire session and synch it real time, including state of run time apps. Not going to happen.

Client side Type 1 is not going to be mainstream for several years. OEMs need to ship first. That’s the only way to force the issue with MS.

VDI is not a cost only play. That should be clear by now. The technology to make it cheap is not here yet. VDI assumes cost savings via single image management. NOBODY at scale is implementing like this or can. Is there anybody who is? Please share how you pulled it off? All your users must love grey and you solved the IOPS problem..... This brings up the same problem we face with TS, that forces it to go niche. And before you say it Appsense, RTO, Sepago, RES, UIA are not the answer. It will take a lot more. If this can be built then hosting providers can make it far cheaper and many of the barriers to adoption AKA CapEx can be removed. Only way to do VDI at scale today is to directly connect and offload some apps locally that don’t work AKA reverse seamless and HDX Connect if you understand that Citrix is the only WAN game in town. And BTW you have to use local disk to spread the IPOS load many many resources just like TS! Ruben S has a great article on this problem of you haven’t already read it.

I agree the Fat desktop as we know it is dead, but it will take a very very long time to die. There are just too many dickheads in the world that will net change their legacy ways or thinking. The best thing this while focus is going to bring is the power of what we already know in SBC land to the old school desktop folks. This will result in a much bigger market overall, but ultimately much innovation is still needed to make it real and personal enough for the masses.