In today’s volatile business world, companies must be agile and efficient. Unfortunately, data centers and server rooms have grown increasingly complex — and many have become unwieldy and difficult to manage.
When companies give each application its own server, running only one operating system, “server sprawl” is the result. Up to 85 percent of each server’s resources may go unused. The resulting excesses in hardware, power, cooling and management can lead to infrastructure instability.
Although downtime is good for vacations, it’s not good for business.
Just ask the IT team at Smead Manufacturing, a supplier of office filing products and records management systems based in Hastings, Minn. An unexpected outage of its legacy PeopleSoft Supply Chain Management (SCM) system cost the firm three days of downtime.
Such a failure would have major repercussions for any business. Fortunately for Smead, a solid solution — server virtualization, and the consolidation and disaster recovery advantages that go with it — ensures that it won’t happen again.
The firm’s first foray into server virtualization occurred about six years ago. Since that time, Smead experienced some network improvements related to the virtualization efforts. But the company didn’t fully buy into the application until the downtime occurred.
“Running a [legacy] PeopleSoft suite of SCM software two versions behind meant that the applications required an older operating system and older hardware technology to support it,” says Albert Lui, the company’s chief technology officer.
“While security patches were being performed on the Microsoft Cluster Server Software [MSCS], the configuration was broken,” he adds. A break in the shared-disk configuration or damaged data on the disks brought the system down for a few days.
Debilitating as it was, the breakdown confirmed that Smead needed updated server technology to sustain its legacy applications in the future.
“Furthermore, when we looked at the [server] utilization report, it confirmed that we were terribly underutilized,” Lui says. In fact, the report showed that 85 percent of servers were running at an average of only 15 to 20 percent capacity.
Immediately, Lui saw server virtualization as the solution. “It would provide us with better utilization of all the resources on the server side,” he says. “In addition, we could support applications on older OSes in a virtual machine to continue sustained applications.”
Given the potential return on investment, Lui and his team won complete buy-in from management, which allocated $1.3 million to complete the server virtualization project.
Smead’s physical-to-virtual (P2V) migration began in April 2008, and the virtualized infrastructure was put into production at the end of September 2008. “We started with 206 physical servers in August 2008,” Lui says. “As of March 2009, we had only 107 physical servers.
“We virtualized 99 physical servers in that time. Now, when new applications come in, we don’t buy more physical machines. We just add more virtual servers to the system. Thus far, we’ve avoided purchasing 48 new servers,” he adds.
The P2V migration was a straightforward process. “Once you put it into production, it goes very fast,” says Klaas Snater, senior network analyst at Smead. “Our original target was to virtualize 45 servers. We completed that within three weeks.”
Once the company determined that server consolidation via virtualization was the best approach, one of the major objectives was to virtualize all commodity servers used to sustain the PeopleSoft suite of applications and other critical line-of-business applications.
“We needed something to sustain use of the [legacy PeopleSoft system],” Lui says. “This was rather than spending millions on an upgrade for the SCM system.”
To the delight of all stakeholders, the project came in under budget by $380,000. What’s more, adds Lui, by eliminating new servers, the company has saved almost $1 million in cost avoidance alone.
“The cost avoidance of new commodity servers within the first four to five months was close to $915,000,” he says. “I don’t need to buy new servers for technology refresh or new applications, and when we retire those servers, I cut down all the maintenance cost of keeping them.
“If I look at operations expansion and increases in application requirements, I figure we could save $3.2 million over five years.”
Daniel Sorenson, lead technical analyst at Smead, points out the company also succeeded in achieving its primary objective: to support older operating systems on newer hardware that was incapable of running the legacy applications.
“We also realized power savings by utilizing fewer servers,” Sorenson says. “As we looked at our redundant uninterruptible power supply systems, we were using the maximum of our UPS load.”
Smead chose to go with a Microsoft Hyper-V Server environment six years ago, when Microsoft acquired Connectix, which later became Virtual PC, then Virtual Server and finally Hyper-V.
“At that time, we had joint ventures with Microsoft and Unisys to do testing on the Unisys ES7000,” Lui says. “We found the virtualization software was not scalable enough, as it was limited to using only one CPU, so we put it aside.”
Because Smead already had a Microsoft platform, the move to a virtualized environment using Hyper-V gave it the best cost of ownership, he says. “That’s why we picked the Hyper-V product above any others.”
Snater adds, “The beauty of Hyper-V is that you have the same [virtual] hardware all over the place, so you don’t have to find different drivers for a variety of servers, and you don’t have to find different products to support the applications. That’s a big win in time savings and troubleshooting.”
Smead now has close to 60 services running in its service catalog — a list of services that the firm provides to staff and customers. Almost 80 percent are now in a virtualized environment.
The P2V migration itself was very efficient and fast, says Snater. “The only thing that we are missing is the capability of going back from virtual to physical,” he says. “Microsoft doesn’t support that.”
But why go back at all? Business needs change, as does the demand for application sizes. And the firm may need more flexibility in the future.
“If a developer doesn’t support virtual servers and they want us to reproduce an application error, you may have to put it on a physical box again without reinstalling everything,” Snater says.
Third-party vendors offer a pathway back to physical servers — at an additional cost. “To be honest, you’re taking some risk in moving some non-Hyper-V supported applications over to Hyper-V,” he says. “But we tested them, and they run. And we haven’t experienced issues. We have a process server that determines whether an application can be virtualized. So we use a third-party virtual-to-physical as a potential fall-back.”
The company also virtualized its Logan, Ohio, disaster recovery site. “We know that disaster recovery is much easier and faster with a virtualized system,” Lui says.
By day, the disaster recovery site is a development test system. But if a disaster occurs, the company can switch the system over to disaster recovery mode and activate a production copy, whether it’s a database or an application.
“We look at this strategy as a method for efficiency,” Sorenson says. “It enables [us to have] a copy of our production virtual machines at our DR site.”
Managing Virtual Environments
Smead implemented Microsoft System Center Operations Manager 2007 to help manage performance and operations. The product works with Microsoft software and applications, allowing for greater control of the IT environment.
The company also uses Microsoft Systems Center Configuration Manager.
“We use SCCM to manage an inventory of all our servers, including the virtual machines,” Sorenson says. “And we plan to implement the Desired Configuration Management module. This is to make sure that all VMs are identically configured.”
Smead uses Systems Center Virtual Machine Manager to provision physical-to-virtual capabilities and to manage the overall VM and Hyper-V infrastructure.
“Now we have a standard image or standard build that’s given us the ability to provision a new machine in 15 or 20 minutes, compared with a whole day before virtualization,” says Snater. “What we’re looking at is the automated provisioning or portal provisioning to be in place within six months.”