Kronos Inc. was suffering from a serious case of server sprawl. Some 330 boxes had pretty much used up the space, power and cooling resources at the workforce management software company's Chelmsford, Mass., data center. The problem became acute around July 2006, when the company acquired another firm -- and 80 more servers.
"We literally had nowhere to put them or plug them in," said IT manager Raymond DeMartini. "That pushed us over the edge," into taking the virtualization plunge.
By deploying multiple applications across virtual machines on a single server box, the company was able to significantly improve resource utilization and achieve a 16-to-1 server consolidation ratio. This, in turn, decreased power consumption by 20% and eliminated 25 racks from the data center. Virtualization has also reduced unplanned downtime by 88%, and resulted in 97% faster server deployment, according to Michael Moran, senior IT systems manager at Kronos.
Moran credited the project's success in large part to a rigorous upfront capacity planning process his team performed with the help of Bedford, N.H.-based systems integrator Expert Server Group. First, his team used PowerRecon from Novell Inc. subsidiary PlateSpin to track application workloads on existing physical servers, over a complete activity cycle. "We found that we had a lot of single-application systems in the data center, whose load utilization, even during peak times, wasn't more than 30% or 40%," Moran noted. About 150 servers were tapped as viable virtualization candidates. Data generated by PowerRecon also helped Kronos determine storage allocations for the virtualized environment.
Other midmarket companies would do well to take a page from Kronos' book, industry sources agree.
With midmarket IT budgets continuing to tighten, "overprovisioning is a luxury midmarket companies can't afford," noted Audrey Rasmussen, a principal analyst at Ptak, Noel & Associates LLC. More and more CIOs are turning to virtualization as a means of getting more bang for their server bucks. However, up-front capacity planning is critical to such projects, to ensure that efficiencies don't come at the expense of performance and reliability, she added.
Even so, too many firms are still managing capacity by waiting until problems occur, "and then allocating more resources (and more, and more) until they go away," Enterprise Management Associates (EMA) noted in its April 2008 report, "Virtualization and Management: Trends, Forecasts and Recommendations." The report goes on to state: "Enterprises … need to look for sophisticated tools that understand resource allocation and usage across the physical and virtual ecosystems, and across the full range of the virtual environment (hosts, guests, hypervisors, clients, storage systems, network components, etc.)."
IT managers need to measure not just overall server performance during peak usage times, but also different applications' demands on specific resources, like I/O ports and memory, at different times of the day and month, said Andi Mann, research director at Boulder, Colo.-based EMA. This enables them to make optimal use of server resources by colocating, for example, a financial application that makes heavy use of the CPU during the day, with an I/O-intensive backup application that does its heavy work at night. It also ensures that "a server's CPU doesn't get instantaneously saturated by multiple workloads making calls simultaneously," Mann noted.
Overprovisioning is a luxury midmarket companies can't afford.
Audrey Rasmussen, principal analyst, Ptak, Noel & Associates LLC
According to Rasmussen, companies also need to monitor performance and manage capacity on an ongoing basis. This is true of any data center installation, but particularly critical for a virtualized environment, where workloads keep shifting among virtual machines (VMs), and VMs among servers.
The good news is vendors are stepping up to the plate. Leading management platform vendors like IBM, Hewlett-Packard Co. and BMC Software Inc. have been aggressively targeting the virtualized space, often through acquisition. Cirba Inc., Novell/PlateSpin, Akorri Inc. and, of course, VMware Inc. offer virtualized capacity management tools.
The not-so-good news is the market remains highly fragmented. Ideally, software tools across key IT Infrastructure Library disciplines like performance and utilization monitoring, capacity and storage management would all feed into and access a common configuration management database, Mann said. However, an EMA study found that less than 50% of virtualization management tools support "cross-discipline integration," he added.
Still, savvy IT executives are figuring out ways to harness complementary management tools, often with the help of systems integrators and the vendors themselves. Kronos, for example, is using Akorri's BalancePoint to monitor utilization and proactively manage performance across a virtualized infrastructure. "When users start complaining about response time, it helps us pinpoint whether the problem is with the FC switch, storage array or VM processes," DeMartini said.
Elisabeth Horwitt is a contributing writer based in Waban, Mass. Write to her at firstname.lastname@example.org.