Don’t Believe Any Numbers You Don’t Make up Yourself

Microsoft, in emphasizing the value of Windows Server, has stated that open-source Linux is "free like a puppy". It now argues the other side against VMware by constantly equating the "free" licensing cost of Hyper-V with value. The comparison fails to consider how the respective platforms stack up in their ability to assist organizations in achieving their ultimate virtualization objectives.

Where's the Beef?

McDonald's founder Ray Kroc liked to claim, "We take the hamburger more seriously than anyone else". VMware could say the same thing about virtualization. Every page of its Web site educates viewers on different aspects of virtualization, but the term is not even mentioned on Microsoft's home page. Navigating to Microsoft's virtualization section reveals the biggest emphasis to be on price comparisons between Microsoft and VMware. Videos, white papers, blogs, case studies, analyst quotes and even Microsoft's ROI calculator tool underline the cost competition with VMware.

VMware's suite of network, management and automation tools enable a virtualized data center, disaster recovery solution and desktops. Microsoft, alternatively, emphasizes its ability to address both physical and virtual infrastructure. While VMware stresses the superior performance, security and reliability of its hypervisor architecture, Microsoft promotes as advantageous that Hyper-V is incorporated as a feature of the operating system.

Unlike an operating system running an individual application on an isolated server, virtualization integrates formerly disparate silos of compute, storage and network resources. Virtualizing a production environment results in zero tolerance for issues in performance, reliability or security that now affect the entire IT operation. While the many unique capabilities of vSphere such as the distributed virtual network switch, storage live migration and integrated VM level fault tolerance may be argued to be essential for a virtualized data center, nothing is more important than reliability. Duncan Epping of yellow-bricks.com fame gave a succinct response when asked why he used VMware: "Because it's rock solid!"

The Microsoft ROI Calculator

Microsoft's Web site audaciously claims that Hyper-V is one-third the cost of VMware. As Table 1 shows, the cost of Windows licensing is factored in when purchasing VMware although this is only a requirement when running Windows VMs. It also disregards the obvious deployment of Windows Server Datacenter Edition which allows unlimited Windows Server instances on an ESX/vSphere host. Microsoft does concede in its "one-third the price" video that the Hyper-V to VMware cost ratio increases to 50 per cent if using Datacenter Edition.

Table 1: From Microsoft Web site showing Hyper-V 1/3rd the cost of ESX (ESXi is a typo)

Microsoft also takes the very unusual approach of using its Virtualization ROI Calculator to compare ROI results with ESX. While an ROI metric can be useful in evaluating the economic impact of different platform investments, all of the relevant variables must be factored for each scenario. The Microsoft calculations erroneously treat Hyper-V and ESX exactly the same.

I ran a virtualization scenario on both the Microsoft and VMware calculators for comparison purposes. Both tools have a similar look and feel which is not surprising since they were created by Alinean, and both show a very similar 3-year net cost reduction from the physical scenario. I used the following parameters, but otherwise Microsoft default figures wherever possible:

  • Analysis period: 3 years
  • 350 total physical servers: 100 2CPU/1 core, 200 2CPU/2 core, 50 4CPU/2 core
  • Virtualization hosts: 2 CPU / 4 core
  • Users: 5,000 (the Microsoft tool requires an input for # of users in order to work)
  • Cost of capital: 0 (for simplification purposes)
  • Cost variables to consider: Existing servers only
  • Virtualization: 100% (this percentage is fixed for both tools)
  • VMs/host: VMware uses a ratio of 8.31:1 (consolidating to 42 hosts), while Microsoft uses a ratio of 5:1, consolidating to 70 hosts (for both Hyper-V and ESX).

The output from the Microsoft ROI tool, shown in Table 2, has a 3-year net cost savings for both the Hyper-V and vSphere scenarios of $2,083,449 – but an initial Hyper-V 3-year investment of $399,801 vs. $1,448,561 for vSphere. Table 3 shows a breakdown of the investment components for each solution.

Table 2: Output from Microsoft ROI Tool with 350 servers, 5,000 users.

Analyzing the Microsoft Calculator Investment Figures

INVESTMENT 

 Hyper-V

 vSphere

Windows Server Ent. Licensing

 $ 88,113

 $    88,113

CALs

 $     

 $    78,400

ESX Licenses

 $     

 $  402,500

vCenter

 $     

 $   5,000

Backup Software

 $     

 $  287,000

Storage

 $ 73,500

 $    73,500

Professional Services/Installation

 $  238,188

 $ 238,188

Training

 $     

 $    10,000

   Total Initial Investment

 $  399,801

 $ 1,182,701

   Total 3-Year Investment

 $  399,801

 $ 1,448,561

ROI (over 3 years) 

521%

71%

 Table 3: Investment Summary from the Microsoft ROI Tool

Windows Server Enterprise Licensing

The Microsoft tool assumes that all hosts run Windows guest OS's, and then calculates Windows Server Enterprise rather than Datacenter because of an assumed server consolidation ratio of 5:1 (which makes Enterprise less expensive). If the ratio is larger than 5:1, it quickly becomes advantageous to utilize Datacenter Edition.

Although the Microsoft tool assumes identical hosting ratios between Hyper-V and ESX, evidence suggests otherwise. Evaluations by our engineers indicate a ratio of about 2:1 in VMs per host under ESX versus Hyper-V. A quick Google search shows this ratio or higher to be commonplace. A ratio ranging between 1.5:1 and 2:1 was the conclusion of a VMware funded study by the Taneja Group. The 2:1 ratio is supported by comparing the internal consolidation ratios achieved by Microsoft as outlined in its January 2009 Technical Case Study as opposed to those achieved by VMware as presented in its January 2009 Webcast by CIO Tayloe Stansbury. A ratio of 1.67:1 is obtained by using the respective consolidation ratios from the VMware and Microsoft calculator tools.

A larger hosting density for VMware significantly lowers both the investment and three-year operational costs by reducing the number of virtualization hosts and their associated expenses. It also reduces licensing requirements for Microsoft products such as Windows Server Data Center Edition and SQL Server Enterprise.

CALs

$80,000 worth of CALs are applied to the VMware investment cost, but not to Hyper-V. I can find no reason for ascribing CAL costs to the VMware scenario, and assume it is a mistake in Alinean's methodology.

ESX Licenses

The "initial" cost of ESX licensing includes three years of annual subscription and support (SNS) at full list, although VMware offers a significant discount for purchasing SNS up-front. A greater hosting density would result in a decreased cost for both licensing and SNS.

Backup Software

In Microsoft's view, each ESX host requires an additional $4,100 in management/backup software to put it at parity with Hyper-V. Although the aggregate cost of $287,000 equates to 20% of the total VMware investment, I was not able to find an explanation for it.

Storage

The cost of storage appears to be significantly understated for typical virtualization situations. The VMware calculator tool reflects an estimated storage cost of $432,000 for the same scenario. VMware says that vSphere capabilities such as thin provisioning now reduce storage costs by over 50%, a claim appearing to be corroborated by testing results beginning to emerge from the field. For an organization with significant storage requirements, vSphere may have a substantial storage cost advantage over Hyper-V.

Professional Services/Installation

Microsoft likes to tout that Hyper-V is based on familiar Windows interfaces and works with well-known Windows-based technologies. Most midsized and larger organizations already utilize VMware meaning that IT is familiar with its technology and management tools. As an organization climbs closer to 100% virtualization, it is able to leverage its staff expertise to produce still greater gains with minimal further investment in training.

Training

The Microsoft tool applies a cost of $10,000 for training to the VMware scenario but not to the Microsoft scenario (though a note confirms it should be applied to both). This appears to be another Alinean error.

Return on Investment

The cost of the respective investment for both the vSphere and Hyper-V scenarios is subtracted from the savings to result in the net savings ($2,083,449 in Table 3), but then the cost for the Hyper-V investment is subtracted again from both numbers when calculating the ROI. This is clearly an Alinean computational error.

100% Data Center Virtualization

Up until now, most organizations have generally been reluctant to virtualize the majority of their production servers. My soon-to-be-published article, Perfect Storm for a 100% Virtualized Internal Cloud, describes how vSphere 4 provides the performance, fault tolerance, dynamic resource allocation, networking capabilities, security, management and API components necessary to successfully obtain 100% virtualization. The greater the percentage of data center virtualization, the more an organization is able to save and consequently leverage its virtualization investment.

VMware says its internal IT will be 100% virtualized by the end of the month in contrast to Microsoft which says it is shooting for an internal virtualization ratio of 50%. Although Hyper-V R2 will resolve some of the challenges Microsoft's IT department faces today, it will still be hampered by limitations such as an inability to apply network and security policies that are live migration aware. Even if a Hyper-V organization could virtualize 80% of its servers as opposed to 100% for vSphere, it would still suffer a markedly higher operating cost over the analysis period. The number of physical servers in a Hyper-V scenario is increased still further if an organization uses Linux other than SUSE.

Does ROI Matter?

Virtualization is a rare IT technology that provides a "hard" ROI without quantifying intangibles such as lowered risk and increased agility, or even IT staff benefits. Consolidating servers by 90% or more slashes associated costs such as power, cooling, cabling, power whips, generator and UPS slices, rack space, distributed network switch ports, core network switch ports, SAN ports, HBA ports and maintenance contracts. Over a period of a few years, these cost savings can easily amount to hundreds of thousands of dollars for organizations with around 25 servers or more, ranging up to millions or tens of millions for larger organizations.

Running varying virtualization scenarios using my own ROI tools all show vSphere as the superior economic choice. But as the title of this blog says, you shouldn't believe any numbers that you don't make up yourself. I encourage IT leaders considering virtualization to not just rely on manufacturer calculator tools, but to take the time to carefully evaluate the economics of different virtualization solutions in their unique environments.

ROI can be very helpful in deciding whether or not to virtualize, but it is typically not a good metric for comparing different products. The huge economic and other benefits resulting from a successfully virtualized data center inevitably make the cost comparison between individual components irrelevant in comparison to their ability to facilitate the desired outcome.

————————————

Follow-up:  Tom Pisello of Alinean wrote a thorough response to this blog.  Alinean seems to agree with most of my points, and is even "posting an update to proactively address" the issue of CALs (this section of their tool was defined by Microsoft).  Alinean apparently did include notes regarding the backup software (I was unable to find them when writing the blog, but will look again when I get a chance), and found no calculation error with the ROI metric (I went through this carefully before reporting it – but since the error was repeated for both Microsoft and VMware scenarios, it wasn't material enough to warrant me running through all the scenarios again to replicate the issue).

Overall, I thought Tom's response was great – nondefensive and clearly representative of a sincere intention to provide organizations with as accurate results as possible.

This entry was posted in Virtualization. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *


*