VMware View is a server hosted desktop virtualization solution that supports Windows-based virtual machines that execute on VMware’s vSphere hypervisor platform. As a server hosted desktop virtualization solution VMware View requires connection between the client workstation and a central server instance that hosts the virtual desktop environment. View also provides “experimental” support for disconnected operation using View Local Mode.
I am looking to make a number of cost models available comparing all forms of desktop virtualization solutions against a baseline set of functional and nonfunctional requirements. Wherever possible I intend to use published reference architectures, modified when necessary to accommodate real-world requirements, that meet or exceed the standard baseline requirements I published here.
My initial goal is to develop a cost model that will allow me to calculate the cost of a mid-sized (500 – 5,000 users) VMware View 4.6 environment, and to this end I’m using the information provided in HP’s “Enterprise Reference Architecture for Client Virtualization for HP VirtualSystem” as my baseline. I don’t really consider this reference architecture model to be suitable for production use due to shortcomings allowing system availability – there is no accommodation for individual component failure nor is there a provision for a business continuity environment – something that I consider a necessity for any production VDI deployments. These two points aside the HP reference architecture is well thought out and lends itself well to developing a basic cost model. Further refinement of the model will be made over time to incorporate additional cost elements such as data center networking hardware, data center power and cooling systems and so on as necessary to implement a comprehensive solution.
FAQ
Where did the costs, equipment specification and sizing data come from?
All hardware pricing except for the HP P4800 SAN was obtained from the HP Small and Medium Business store front. The P4800 costs are not available from the Web store and were obtained direct from Kate Davis (@kateatHP) the HP P4000 product marketing manager.
The pricing and other data points used in the HP VirtualSystem for VMware View cost model were obtained as follows:
- Reference Architecture - HP Enterprise Reference Architecture for Client Virtualization for HP VirtualSystem
- Modifications needed to support production deployment - myself
- VMware View Costs - VMware
- Windows VDA Costs – Microsoft – Licensing Windows for Virtual Desktops
- Data Center Power and Cooling Costs – Estimated cost provided by subject matter expert
- Server Power Consumption - HP Power Advisor
- Storage Power Consumption - HP Power Advisor
- Network Switch Power Consumption - Estimate provided by subject matter expert
Are the published prices realistic?
That is an extremely difficult question to answer. The price is that HP publishes through the HP Small and Medium Business store front are real, credit card in hand you can logon and buy equipment at these prices. However, once volume enters the picture, substantial discounts can be obtained. what is substantial? again that is hard to answer, the size of discount that may be achieved depends on the size of the order, local market conditions, when in detailed here you are buying, and the marketing value that a vendor might consider to having you as a customer. The same applies for VMware View licenses. There have been many stories suggesting that VMware has offered very generous deals to View customers, however I do not have any first-hand knowledge of this, and must treated with a certain amount of skepticism. Microsoft Windows licensing is also subject to discounting, although, I perhaps not as generously as VMware might offer.
The pricing for storage, is more problematic. Again large discounts can be obtained but few organizations would consider the P4800 SAN for larger deployments, and this inevitably means that there will be pricing inaccuracies here. However it has to be noted that the P4800 is considered to be a low-end solution with consequent low storage costs in comparison to some other vendors solutions.
What is the “Oversubcription” value for?
Oversubscription is the percentage by which you can override the cost model’s server sizing rules. For example, if a VDI server supported 80 concurrent sessions without an oversubscription setting sizing a system to support 81 sessions would automatically require two servers. Within oversubscription of 5% the cost model will allow a server to be oversubscribed by four sessions before allocates a second server. Setting an oversubscription value of zero will rigidly enforce system sizing rules.
Why were thin clients used when existing PCs could be retained?
There are many justifications for replacing PCs with thin clients as part of a server hosted virtual desktop deployment including:
- Longer life
- Lower capital cost
- Lower support and administration costs
- Lower power consumption
- Improved data security
The default settings of the configuration switches ‘Include desktop PC refresh (Y/N)” etc. were set based on the following assumptions:
Include Hot Standby Servers – Y
All server hosted virtual desktop systems require continuous availability of virtual infrastructure servers for services to be maintained. Failure of one or more servers without spare capacity being immediately available will compromise performance especially during times of unforeseen demand.
Include Data Center Network Switch Ports – Y
In many data center environments individual projects are required to fund their own network infrastructure hardware, although blade server environments require few ports individual port costs are high and need to be taken into consideration.
Include Business Continuity Servers – Y
As with Hot Standby Servers; Aal server hosted virtual desktop systems require continuous availability of virtual infrastructure servers for services to be maintained. In the event of data center outage all virtual desktop services will be lost unless business continuity servers are provided.
Include Data Center Power and Cooling Costs – Y
Server hosted virtual desktop systems require significant data center power and cooling to support virtual infrastructure servers and storage systems.
Replace PCs with Thin Clients – Y
As indicated above there are many good reasons to recommend replacing desktop PCs with thin clients in VDI environments; however, there are circumstances where it may be desirable to defer replacement (budgetary) or functional (maptops for mobile users).
Note: the configuration switches can be adjusted to meet individual needs and the cost model will recalculate accordingly.
Will you be providing an operating costs model?
Yes. Creating a general OpEx model will take some time, but in the interim I can assist with developing models for specific environments.
What if I disagree with your assumptions?
You are free to change any of the input fields, for example if you have obtained better pricing for any hardware or software component, if your test results show you can get more or fewer sessions per server.
Aside from that, please feel free to share any comments you may have or make recommendations for improvement. For my part, I will endeavor to respond to all constructive comments and incorporate any changes that improve the accuracy of the cost model.
Can I have a copy of your cost model?
I’m a firm believer in the value of open source projects, and as you can see haven’t protected the cost model calculations in any way. So I’m not going to stop you copying it other than to draw attention to the license which permits it use for non-commercial purposes only. Having said that I would prefer it if you could link to this page rather than embedding the cost model in to your own page or copying it; partly because I’d like to be sure that anyone using it has access to the latest version of the model and any accompany in notes, but also because I think there’s more value to be had by cooperating to improve a single model than in creating multiple competing models.
If you are interested in using it for commercial purposes, then please get in touch.
Change Log
r1.3
Formatting improvements
New feature – Introduced a new configuration switch that supports replacement of a percentage of desktop PCs with thin clients, along with a data entry cell to input the percentage to be changed.
Updated FAQ to reference this change.
r1.2
Formatting improvements
Addition of data center networking port calculations – number of ports required, costs, and power calculations. Network Port calculations
r 1.1
Corrected error in Windows VDA license calculation.
Addition of data center power and cooling calculations – calculation of power consumption by data center servers and storage and calculation of cost of provisioning power consumed.
r 1.1
Initial Release
Developed cost model using the information provided in HP’s “Enterprise Reference Architecture for Client Virtualization for HP VirtualSystem” as baseline. With appropriate modifications to support real world production environments i.e., N+1 redundancy for all infrastructure servers and business continuity to provide continuous operation in the event of data center outage.
Hi Simon,
Can you please specify the exact configuration of BL490C and BL460c Please
Simon, in your experience, what would the typical fee for professional services be to install, and deploy a 1,000 seat VDI instance?
A few additional points to consider:
DR: N+1 failover is possible, with one critical assumption: network capacity between sites. Granted, most “enterprise class” clients will have this but a geographically distributed mid size company is unlikely to have this network capacity. Not sure how the Cost Model can account for client type or network capacity. Distributed N+1 also adds a significant Opex cost for additional staff and data center facilities. Having said that, this model does focus on Capex.
Mobile: means off line. And that usually means disconnected from the network a good portion of the time, or at least limited to Starbucks WiFi (I may be injecting personal work habits here).
Using a remote desktop display technology does not solve the problem of mobile. This is why most companies struggle with VDI. VDI is an excellent tool until you need to go off line.
Graphics applications: Most of the apps we use day to day rely on graphics. Plus, VoIP is a staple tool for many mid size companies. Both of these are contra indicated for VDI or any remote desktop protocol. Yes, there are lots of tools to improve or accelerate graphics, network and VDI performance, but they add significant cost and complexity.
Professional services: I do not know a single enterprise client with the internal skills to successfully deploy VDI. Hence, the capex for Professional Services must be added to the cost model. Some firms may account for this as an Opex item but most do not. What does it take to deploy a 1,000 seat VDI site – servers, rack and stack, networking, storage, virtualization, etc etc. Would that be between $100k-250K, plus cost overruns and follow up tweaking.
Licence costs: Microsoft licences their products by end point device when hosted on a server. Most VDI analysis underestimate the number of devices required and thus the Microsoft licence costs.
It’s interesting that Microsoft has published a white paper ” VDI TCO Whitepaper for Office Workers” comparing VDI to a “traditional” Windows 7 installation, and claim that W7 is less expensive by about 11%. Not saying I agree with their analysis, but they do make a compelling argument. And lets not forget, Microsoft has the “ear” of many CIOs.
I was recently at a conference where someone asked the question; “Why are so many VDI projects failing, and why do so many VDI project managers get the chop?”
I think the reasons are simple: VDI is “sold” based on the simplest use case; low end task workers. And it works fine for them. But, as we expand the scope to remote office workers and mobile knowledge workers, the costs and complexity go up, and the performance and user satisfaction go down.
If you consider a typical 1,000 employee company, maybe 30-40% will be in a LAN connected head office while the rest are in remote offices or are mobile. Easy to see how a VDI project can stall, and fail to deliver the hoped for TCO and ROI.
Minor correction suggestion. When electing to NOT replace PCs with Thin Client, the Thin Client capital cost should be eliminated. Easiest way to fix this is to remove the question and only use the percentage (with a default value of 0%).
I agree with Derek’s comments about DR (and realize that you have included disclaimers). One approach is to use C3000 enclosures in local facilities with a data backup and DR failover plan in an N+1 configuration (multiple sites failover to a common single site). This avoids the 2N configuration required of conventional dual-site DR for centralized VDI. It also eliminates any WAN latency concerns.
I agree with Derek about the need to support mobile platforms. It seems he is suggesting that an opex benefit be considered for avoiding the data consequences of a robust laptop. If you decide to build an opex model to accompany this capex model, that factor should be included. In my opinion, the cheapest way to support fractional deployment of mobile extensions is to use remote control of the user’s primary desktop.
Consider replacing the SB40 blades with D2200sb blades.
Do you need any discussion about profile management or application streaming (e.g. disclaimers)?
Are you presuming static assignment of users to blades? (e.g. If my blade fails, I am down until someone fixes it?)
Another issue to consider with the cost analysis is that most “knowledge workers” are mobile.
An April 2009 report on telecommuting authored by Forrester Group states “German enterprises reported the highest portion of employees away from the office traveling on business more than four days per month — 30%, compared with an average of 22% in North America, 18% in the UK, and 14% in France”.
So, at a minimum, the cost analysis should include 22% mobile computer use (assumes North America). However, a Ponemon report from April 2009 commissioned by Dell found that 31% of employees were issued a laptop as their primary computing device, and that this would grow to over 60% within five years. That means that at least 220 laptops and associated Microsoft licences should be included in the model. Most firms delivering thin clients also provide a laptop for travel.
If we say the average “business quality” laptop is $1,500, then we need to add at least another $330,000 to the capital cost. Plus, of course the additional support and most importantly, security issues associated with data loss due to lost or stolen laptops.
Lost laptops are a major issue, and cause lost productivity and loss of critical, proprietary and/or confidential data. This can result in significant fines (i.e. loss of public data in Massachusetts can result in fines that are up to $5,000….per lost record, fines and penalties for data loss under HIPAA are extremely punitive).
How big is the lost laptop issue? Again, according to a Ponemon study performed for Intel and Dell, 637,000 laptops were lost or stolen in the top 35 US airports in 2009. The study found that over 50% of US companies had lost critical data due to lost laptops, and that cost to recover data was on average marginally less than $50,000 per laptop.
If we are to consider thin client computing for knowledge workers, we must recognize how they work, and the costs and risks that these work habits incur. To exclude these true computing costs is to ignore the realities of today’s market and makes the cost analysis model all but meaningless in real terms.
Simon, as before, I applaud your efforts to build a model that captures the realities of desktop virtualization.
May I suggest a scenario where this analysis may be lacking in “real world” use cases.
With legacy desktops, if a single PC fails, one user is down. No big deal for the organization. If the VDI infrastructure fails, all users are down. That is a big deal!
So, lets consider a disaster scenario….hurricane, tornado, earthquake, fire, flood, ice storm….all of which we see regularly in North America. And of course, mundane power outages or equipment failure.
The frequency of these events suggests that a fundamental risk management issue has been overlooked in this cost model (and every other cost model i have seen).
If we assume only a 50% redundancy, what impact does this have on the infrastructure configuration, cost, and data center space requirements?
Please note that in the case of VMware, their desktop virtualization solution does not work with their Site Recovery Management solution, hence no DR capability.
Granted, most large enterprises (Fortune 1000 type of companies) mitigate risk with off site DR, and do not house their physical data center in the same physical facilities as their workers. If the data center is put off line, the company can switch processing to another site. This assumes a fully functioning DR plan that has been deployed, tested and proven to work.
But, most small medium businesses (27 million in the US alone) host the bulk of their staff in the same building as their “data center”. Or should we call it the “server room”. Hence, if a disaster were to occur, and lord knows, they seem to happen with ever increasing frequency these days, not only would server work loads be down, but all desktop workloads would fail too.
If the client has virtualized their servers, they may have an offsite DR plan for server workloads, but these are not much use without fully functioning desktops.
Is this a real concern? According the the HP report “Impact on US Small Business of Man Made and Natural Disasters”
A Company that experiences a computer outage lasting more than 10 days will never fully recover financially. 50 percent will be out of business within five years.”
An estimated 25 percent of businesses do not reopen following a major disaster
70 percent of small firms that experience a major data loss go out of business within a year.
Of companies experiencing catastrophic data loss:
• 43% of companies never reopened
• 51% of companies closed within 2 years
• 80% of companies that do not recover from a disaster within one month are likely to go out of business.
• 75% of companies without business continuity plans fail within three years of a disaster
• Companies that aren’t able to resume operations within ten days (of a disaster hit) are not likely to survive.
• Of those businesses that experience a disaster and have no emergency plan, 43 percent never
reopen; of those that do reopen, only 29 percent are still operating two years later.
The report continues with facts and figures that paint a rather bleak picture of what happens to the ill prepared company. The economic effect and loss is staggering.
If we are to consider a significant change to desktop deployment and architecture to enhance business productivity, should we not consider these scenarios and include them in our models?
The final paragraph in the HP reports says it all:
“There are many resources available to help a small or growing business make the proper preparations. Disasters are going to happen. The best way to survive is through preparation. And the best way to prepare is to understand that this can happen to anyone, including the person reading these words.”