Server Virtualization: 6 Management Myths
During the past three to five years, when most companies began rolling out virtual servers in earnest as replacements for physical servers, acceptance has grown so quickly among both end users and IT staffs that more than half of all companies now deploy new applications on virtual servers by preference, rather than physical ones, according to a December study from IDC.
By 2014, 70 percent of all server workloads will run on virtual machines, IDC found.
Unlike the early days of virtualization, when business unit managers often questioned IT decisions that removed the physical server on which their key applications ran, virtualization raises few questions even among frontline business managers, some IT leaders say.
"I don't know there was a lot we had to keep explaining," says Brad Thompson, director of infrastructure engineering at Target, who is leading a three-year effort to reduce seven physical servers in each of the company's 1,755 stores to two, each running virtual machines on Microsoft's Hyper-V hypervisor.
"Our end user community doesn't really want to spend a lot of time worrying about infrastructure except to keep track of what it costs," Thompson says. "They ask if we've piloted it, have we proved it. We show them not only did we do that, we did it in 10 of our highest-volume stores over the holidays. They accept that."
Virtualization's cost savings are so dramatic and the technology is so widespread it's hard for end users to justify concern about having a physical server they can touch, even if they'd prefer to have one, according to Gary Chen, research manager for IDC's Enterprise Virtualization Software business.
Still, some major misconceptions about server virtualization remain, even among companies with large virtual-server installations.
1. You don't know what kind of virtual servers you actually have.
An upcoming IDC survey of CIOs and senior-level IT managers shows clearly that, among companies that have virtualized large chunks of their infrastructures and are using the VMs for critical applications, few are using hypervisors from more than one vendor, Chen says.
"About 70 percent of people said they're sticking with a single hypervisor for the time being," Chen says. "About 30 percent said they're open to using two - not 30 percent that are doing it, 30 percent are open to it. About half of those, 15 percent of the total, said they'll probably have more than one hypervisor in a year or two."
But many of those survey respondents are either kidding themselves or don't spend enough time in their own data centers, according to Dan Olds, principal of Gabriel Consulting Group, who has been doing annual surveys of Windows and Unix-server users for more than five years.
"The surveys I do are more with data-center people than CIO or upper-IT management, so their view of what is going on with their servers is a little more hands-on than someone who may have cut the enterprise deal with VMware," he says.
Olds' surveys show VMware is the leading hypervisor, just as IDC's do, but the single-vs.-multiple hypervisor percentages are mirror images. "Consistently, across all sizes of companies and industries, about 71 percent of companies say they have more than one type of virtual server," Olds says. "Citrix and Hyper-V are inexpensive and easy to get, so there's a lot of tire-kicking going on, but it's also a grass-roots thing like Linux was, where things are being used in the data center beyond what the CIOs may even really know about."
2. That big ROI will drop off.
Virtual servers continue to work more efficiently than physical servers, and make better use of the hardware that's already bought and paid for. But the huge return on investment many companies get when they first convert P to V in one department or data center - and usually repeat in each stage of a multi-phase migration - does not renew itself every year or every time the company has to replace its servers, according to James Staten, vice president and principal analyst at Forrester Research.
"When you're doing lots of physical to virtual migrations you're getting massive cost savings. When you get beyond the point of taking lots of servers out, that drops off," Staten says.
Business units get used to the high payoff from a drop in capital costs, and usually don't pay much attention to the portion of the operational budget that often has to rise to accommodate virtual infrastructures, Chen says.
The dropoff in ROI and increase in operational costs that happen when IT finishes the last stage of a phased-migration to virtual servers and begins simply to support them can make IT look as if it's suddenly costing the company more money, Chen says.
"IT ends up having to explain, hopefully ahead of time, why support or ops costs might go up," he says.
3. Licensing costs can multiply.
Virtual servers are cheaper to move and store, but they're still servers and still cost as much as any other server for software, operating systems and management, Staten says.
In fact, even licensing costs can get out of control for companies that go hog wild on the power and flexible capacity virtual servers can deliver by launching more servers and application instances than their budgets will bear.
"If you've paid for 500-servers worth of one license, and used 600 and the trend is toward a lot more, you see a big bill coming," Staten says. "There is a lot of renegotiating of enterprise license agreements going on for that reason."
The volume of servers involved was a key reason Target decided to use Microsoft's Hyper-V as its primary virtual-server product, rather than VMware."Cost actually is a big part of the reason we opted to go with Hyper-V," Target's Thompson says. "Every store has its own, relatively small footprint, and costs involved with the applications and servers. If you multiply it by 1,755 stores, that becomes a very large number quickly if you don't pay attention to controlling costs."
Even after the money saved from replacing seven physical servers with two, paying additional licenses for the Windows licenses for two host machines and the OS on each VM, multiplied by almost 1,800 stores, is a big bill, though Thompson wouldn't say how big.
Running Hyper-V, which comes free as a feature in Windows Server 2008, "saves us millions in operating expenses per year, and it uses the same management tools, with the same skill sets, we use already," Thompson says.
"Every server has a base license - for Windows or VMware - and another for every guest OS and application," Chen says. "It's virtual; it's not free."
4. Management chores will change, not evaporate.
One of the major theoretical benefits of virtual servers is that they're easier to manage than physical servers because they can be launched, reconfigured and monitored remotely more easily than physical servers, reducing the workload and size of the staff required to maintain them. "That's the assumption, but what our surveys show is that the number [of respondents] saying virtualization has made management easier has been stuck at around half for three years," Olds says.
Specifically, according to Olds' data, only 52 percent of x86-based virtual server users said VMs made their overall server-management load lighter, while 25 percent say VMs made server management harder. Almost as many, 22 percent, aren't sure whether management is easier or harder.
"What's not clear is whether that's because it gets more difficult to manage the servers when customers jam more and more workloads on the same systems, or if the management tools that are available aren't up to the job," Olds says.
5. Managing performance will require a wide range of IT skills.
Perhaps the biggest problem in virtualized deployments is that sometimes, even the IT people responsible for them don't really understand how to set up the infrastructure so it's easier to manage, or manage capacity to the detailed level required, according to Patrick Kuo, a Washington, D.C.-area consultant. Kuo has helped build web and virtual-server infrastructures at Dow Jones, the U.S. Supreme Court, the Defense Information Services Agency and, most recently, D.C. political-news site The Daily Caller.
When physical servers or server farms run into performance problems, IT managers can run in a higher-bandwidth Internet connection or add servers - an approach that limits itself because of the cost and logistical difficulty of adding new physical resources, he says.
Virtualized infrastructures don't have those limits, so the IT people designing the virtual infrastructure have to divide data-center resources in ways they may not have done before, if their backgrounds are limited to systems management rather than including networking, applications management or other disciplines.
"With Daily Caller, and with some other clients we've been able to get better performance with a four-tier architecture that puts a layer of caching first, and the app servers, the web servers behind that, and a replicated database backing them up. The fourth tier is a non-production layer for the editors or content contributors to work with," Kuo says.
N-tier architecture is a basic part of data-center and enterprise-application design, but is often not part of the background, responsibility or expectations of IT managers responsible for virtualization.
Looking at server or application performance as standalone metrics, rather than as an indication of how the whole data-center IT stack functions, puts too much emphasis on the performance of the hardware, rather than managing resources carefully to get the best performance out of the hardware available.
"People tend not to think of tuning in virtualization, but especially in enterprise apps you have to get down into the details. You have to see how the instances are performing under specific workloads and look at the trends and do some analysis on what would happen if your utilization jumped from 80 percent to 90 percent and add resources to the layer that needs it," Kuo says.
6. Cost isn't the only thing that comes with a virtual multiplier.
Not every virtual server is the same, but everything you do to one virtual machine, you generally have to do to all the others, according to Phil Hochmuth, program manager for security products at IDC.
The normal way to monitor physical servers is to install a lightweight software agent that will collect and report performance data to the main console, he says.
With virtual servers, that means installing one agent on the physical server and one each on each VM, he says.
IDC's December report on server virtualization predicts an average of 8.5 VMs per server by 2014. That means not one agent running on each physical server, but 9 or 10, each of which uses additional resources, each of which requires an additional license.
"Even besides the performance issues, you don't want a multiplier on the cost of whatever you're using to manage those servers," Hochmuth says.
Follow everything from CIO.com on Twitter @CIOonline, and the CIO.com Facebook page
Read more about virtualization in CIO's Virtualization Drilldown.