Virtualization can deliver dramatic benefits for data centers, but it can also stress the underlying support infrastructure. Power and cooling systems—which may have been quite sufficient for pre-virtualization needs—could easily become inadequate when data center performance patterns are radically altered. This paper describes some of power challenges related to virtualization—and the readily available technologies to address them.
Is your data center ready for virtualization?
Important power considerations for virtualized IT environments
By Chris Loeffler, Global Applications Manager, Distributed Power Solutions Eaton Corporation
Abstract Virtualization brings the potential to deliver dramatic savings in terms of server count, footprint, power consumption and cooling requirements for data centers. For all its advantages though, virtualization also brings some unique challenges: . Overall power consumption will be lower, but it will be highly variable. . There will be fewer servers, but each one will be more critical than ever. . Applications can be dynamically reallocated at will, but the support infrastructure cannot do the same. . Data center footprint will be smaller, but overall efficiency might still be suboptimal. The power and cooling infrastructure-which may have been quite sufficient for pre-virtualization needs-could easily become inadequate when data center performance patterns are radically altered. The good news is that there are practical and affordable ways to address these challenges and improve data center efficiency in the process. This paper looks at some of the power-related challenges and the readily available technologies to address them.
Contents The hidden challenges of virtualization....................................................................................... 3 The challenge: Each server will be more power-hungry than before. ........................................ 3 The challenge: Each enclosure will be more power-hungry than ever....................................... 4 The challenge: Higher rack densities may exceed available UPS capacity............................... 6 The challenge: Applications and their processing demands can shift at will.............................. 8 The challenge: Overall data center efficiency metrics may still not be what they should. ......... 8 Closing thoughts............................................................................................................................ 9 About Eaton.................................................................................................................................... 9 About the author ............................................................................................................................ 9
WP09-06 www.eaton.com/powerquality March 2009 - 1
Is your data center ready for virtualization?
Important power considerations for virtualized IT environments
A North American Fortune 500 company was facing challenges that look familiar to many data center managers. Server sprawl had led to an inventory of more than 5,000 distributed servers, most of them under 20 percent utilized. If nothing changed, the data center would soon outstrip its power and cooling capacity. A new facility would be required at a projected cost of more than $10 million. Instead, the company turned to virtualization. Mission-critical applications were redeployed onto 150 virtual Linux servers, occupying 20 percent of the original footprint. That left enough raised-floor space to triple processing capacity in the next three years, while cutting support costs in half and trimming $15 million from total cost of ownership.
If these results sound good, they ought to. Virtualization has the promise to maintain or increase computing power and data center performance while controlling costs and extending the value of existing data center facilities. Virtualization enables partitioning, whereby a single physical server runs multiple virtual machines, each with its own independent and secure application and operating system. Instead of provisioning a single, physical server with enough spare (often idle) capacity to support the peak load of a single application, you can now dynamically match available processor power across the data center to meet the variable demands of application workloads. The same work gets done, but there's far less idle capacity. Consider a typical small company, running 240 applications, one application per server, each server operating at about 10 to 15 percent CPU utilization. In a typical scenario, about 200 or more of these applications would be candidates for virtualization, at an average rate of four applications per server. Consolidation would trim the configuration down to 52 physical servers in a virtual... [download for more]
