Sorting through some notes from the Uptime Institute's recent IT Symposium in New York and wanted to share some comments and observations from the presentation made by Steve Sams, who is the company's vice president of global site and facilities services.
Sams has been behind the massive consolidation of IBM's data center capacity over the past few years. These are the sites that IBM uses to support its own internal IT concerns, no necessarily the ones used for outsourcing projects. Under his watch, the company has shrunk 235 locations down to 12. This has had the net effect of eliminating $4.1 billion in operating expenditures from IBM's budget. Yes, $4.1 billion, an amount that is the entire revenue of 4 or 5 "midsize" businesses.
Here are his primary pieces of advice for any IT manager or CIO facing the same dilemma—how to keep increasing IT capacity without requiring more power or physical real estate.
- Extend what you have: It has been well documented that virtualization technology is a huge ally in the quest for data center energy efficiency. In just one data center (a site in Lexington, Kentucky), Sams says IBM was able to eliminate five out of every six servers because the existing ones were underutilized. After the project, however, the company had roughly 8 times the computing capacity than when it started. Some interesting utilization stats, by the way. Sams say most companies are running their Unix servers at about 20 percent capacity, while Wintel servers are usually at less than 10 percent. Mainframes, surprise, surprise, are usually runing at about 95 percent utilitization, he estimates.
- Think modular: By Sams' estimate, roughly 60 percent of the capital costs associated with building out a new data center come from the energy-related components (the distribution system, the cabling, the cooling systems and generators, chillers and so on). Rather than guessing at what capacity you MIGHT need down the road, he suggests thinking modular about these components, so that as the servers scale so, too, can the infrastructure to support them.
- Remember your operating expenses when planning: Because of the density associated with today's server infrastructures, companies should expect to spend roughly $5 in operating expenses for every $1 dedicated to capital expenditures.
IBM applied many of these principles to a data center consolidate for a bank in China. When it was done, it had rationalized 36 sites into just two. It has saved the company roughly $180 million in annual operating expenses (for software, maintenance and power) as a result.
My final comment is a bit of a random rant, because I've been writing about technology for a very, very long time. Just when did "virtualize" become a verb? My spell checker goes nuts every time I use it. Note to Microsoft: If you want to be a virtualization powerhouse, you might want to fix that. :)