VMware claims its virtualization products are cheaper than Microsoft's when the total cost of ownership is calculated over a two-year period. To get to that conclusion, it sponsored a study by Principled Technologies that takes a close look at two years of administrator time and expense required to run the respective systems.
Readers of this report: beware. VMware got itself in trouble last July by charging for vSphere 5 Enterprise Edition based on the amount of virtual memory the customer used. It initially set a 48-GB virtual memory limit per license; three weeks later, in the face of customer feedback, it raised it to 96-GB per license. This study is part of VMware's continued response to that blowup.
Setting the price based on virtual resources instead of physical resources struck directly at the budgets of some of VMware's largest customers. The companies exceeding the 48-GB limit were most likely the skilled implementers gaining the most value from VMware products. Proliferating cores per CPU were constantly raising the number of VMs that could be run on one host. VMware decided to shift the basis for pricing so that it too could ride the CPU escalator. Customers protested vigorously. Microsoft, still pricing per physical CPU, started referring to the VMware "v-tax."
What's a dominant virtualization vendor to do? Well, for one thing, it commissioned a study that shifts attention to the cost of the people required to run virtual machines in the data center.
[ Learn more how VMware vSphere and Microsoft System Center compete. See Microsoft Makes Cloud Management Move On VMware. ]
Principled Technologies compared five administrative tasks run on each vendor's central virtualization system: vSphere 5.1 and vCenter for VMware, and System Center with Virtual Machine Manager for Microsoft. The five tasks and their conclusions were:
1. Move virtual machines to perform routine maintenance on a host server, put the server in maintenance mode, then move the VMs back. The claim: vSphere took 79% less time than System Center.
2. Add new storage volumes to a virtualized cluster using VMware Storage Distributed Resource Scheduler. System Center lacks an equivalent feature, so the testers used System Center's Virtual Machine Manager component, plus an administrator's manual decision-making. The latter included time for a System Center admin to warn users there would be a brief VM application outage. The claim: The operation took "95% less time with VMware as compared to Microsoft."
3. Some VMs engage in intensive I/O, becoming a "noisy neighbor" and interfering with the operation of neighboring VMs on the same host. Virtualization admins like to isolate such a noisy neighbor to protect other users. The vSphere admin can do this through the vCenter management console. Again, System Center has no directly equivalent feature. The claim: It takes 97% less time to do it the VMware way, since it's more automated.
4. Provisioning new hosts is a routine task accomplished by vSphere Auto Deploy and System Center's Configuration Manager 2007 R3 bare metal deployment task sequence. The claim: Executing the provisioning was "up to 78% faster" under vSphere.
5. Perform a non-disruptive disaster recovery test. VMware's Site Recovery Manager was tested against Microsoft's equivalent process. The claim: The test is 94% less time consuming under VMware.
The testers assumed these tests would be carried out several times over a two-year period by a senior systems administrator, paid $88,600 a year. The resulting savings to the customer who chose VMware would be $37,540 in management time costs. Because VMware also operates hosts with a higher concentration of VMs, the study added in capital cost savings, using VMware's Virtualization Cost Per Application calculator.
Here are a few red flags to note, in addition to VMware's sponsorship.
First, the Virtualization Cost Per Application calculator is a black box. Wouldn't it be better to stick to comparisons where neither side is doing out-of-view calculations based on its own assumptions? To its credit, the Principled Technologies testers acknowledge that the calculator assumes "a density advantage of 50% more VMs for VMware over Microsoft." The testers scaled that back to "a more conservative estimate of 25%" and did cost estimates based on that.
Unfortunately, we still don't know whether 25% is accurate.
Second, and most notable to me is the noisy neighbor calculation. The VMware virtualized data center tends to be used by a larger company that's more intensely virtualized than the Microsoft Hyper-V user. For that reason, I suspect noisy neighbors tend to occur more frequently in the VMware environment than in typical Microsoft ones.
I'm also suspicious that an administrative function available in vSphere, for which there's no equivalent in Microsoft's environment, was twice chosen for time-sensitive testing. The automated procedures in vSphere were twice pitted against manual ones. If Microsoft is used more frequently in small and midsize business (SMBs), it's quite possible noisy neighbors occur less frequently there and will not result in the amount of lost administrative time the comparison suggests.
Third, the addition of the storage task, with time injected for the System Center admin to warn application users of an outage, also favors VMware, perhaps excessively so. Yes, it would be a lot more time consuming to proceed in that manner, but that's a good reason why the addition might occur after hours or on a weekend in a Microsoft shop. If that's not optimum, it's still the way many shops function. So I wonder what a more realistic comparison might yield on that count?
By the time you're done reading the five test results, you see through the VMware-centric point of view, and few Windows Server and System Center users are going to be convinced by it. On the other hand, if you're a VMware user, you now have a counterargument if the CFO's been pressing for a reduction in those growing VMware bills--that VMware costs you more up front but saves you administrative time after that. It's hard for VMware defenders to quantify the VMware advantage under the pressure of daily operations. The Principled Technologies study has its flaws, but it's ammunition for those who find little at hand in the pricing argument.
And therein lies the advantage of this study to VMware. It won't convert Microsoft users to more efficient VMware administrative tasks. It's a buttress for VMware's own advocates inside the enterprise, who are under pressure from the rising cost of the VMware share of the budget. At some of VMware's largest customers, the move to virtual memory pricing in July prompted a two-fold reaction: Customers decided to pressure VMware to raise the limit to where they were less affected; they also decided to convert parts of the data center to another set of products to gain leverage over VMware. Microsoft, with "free" virtualization embedded in Windows Server 2008, was an obvious candidate.
If this study had been done from the point of view of the typical Windows administrator, who must manage both physical and virtual resources, it might have reached entirely different conclusions. But, of course, displaying an understanding of System Center administration wasn't the goal. Highlighting the differences between the two, where VMware might claim advantage in a slender band of advanced virtualization characteristics, was the point.
From thin provisioning to replication to federation, virtualization options let you reclaim idle disks, speed recovery, and avoid lock-in. Get the new, all-digital Storage Virtualization Guide issue of Network Computing. (Free registration required.)