December 16, 2008, 3:48 PM — Global corporations must often absorb additional data centers because of mergers, acquisitions and other factors, but too many IT organizations have failed to implement policies that govern how to best align new data centers with current operations, Gartner analysts said Tuesday as the analyst firm's annual data center conference kicked off in Las Vegas.
Â Â Â Â Â
Although trends point to industry-wide consolidation of data centers, many businesses find themselves with more then they used to have due to mergers and acquisitions, various business initiatives and growth, regulations, and need for more capacity, space and energy. Â
A typical enterprise performs occasional data center consolidation projects to deal with whatever changes it has undergone the previous few years, said analysts Donna Scott and Paul McGuckin during a keynote address on data center planning. But enterprises should instead treat data center planning as a continuous process, regularly reviewing a policy and applying it to business events as they occur, they said.
"The right number of data centers and strategy should be your decision," Scott said. "It shouldn't be something that just happens over time."
Forty-two percent of the several hundred IT users at the session operate three or more data centers in North America, according to an instant poll conducted by Gartner. Forty-five percent are expanding or planning to expand data centers in the next two years, while 43% are consolidating or planning to consolidate in the next two years. (Twelve percent didn't know their companies' plans).
In another instant poll, Gartner found that 11% of attendees do not have a disaster-recovery strategy. Â
Scott and McGuckin urged attendees to outline three to five tiers of IT service criticality -- in other words, choosing which applications need the most reliable and fastest IT services and which need something less.
"Not every application needs the same kind of processor or the same kind of facilities," McGuckin said. "We routinely see folks who have built out a highly redundant and expensive infrastructure, and many of the applications running in that data center don't need that level of uptime and redundancy. So that's wasted money. You want to match level of redundancy and uptime to needs of an application."
It's important to know what level of latency each application can tolerate, given that organizations are faced with deciding whether to site data centers in America, China or Europe. Companies that want to go overseas to reduce costs may be forced to stay local because of performance requirements.
"The speed of light is not our friend when it comes to figuring out how to provide the maximum uptime," McGuckin said.
An enterprise's data center plan should define the facilities that host IT services and the strategy for placing those services, taking into account the importance of each application and plans for responding to outages, data corruption and other problems, Gartner says. For global corporations, the strategy should also take into account the different laws in each country regarding intellectual property, privacy, security and labor.
After developing a strategy, an enterprise must identify where the current architecture falls short and come up with migration plans, which might involve expansion or contraction depending on each organization's specific needs.
Enterprises can take many steps to lower the cost of data center operations. Here are a few specific pieces of advice offered by Gartner:
-- Perform an IT service cost analysis before deciding which facility or facilities best suit your needs.
-- Rationalize and consolidate data centers.
-- If latency is not a major concern, host the application or service in a single data center for worldwide access.
-- When latency is a concern, host applications in local centers in each region.
-- Aggressively pursue virtualization.
-- Decommission orphan servers and storage.
-- Replace servers and storage that are five years old or older.
-- Move test systems and half of Web servers from UPS (uninterruptable power supply) to grid power.
-- Use power-capping techniques and cold-aisle containment.