Million-Server Data Centers? Welcome to the Era of Extreme Integration

“This is the era of extreme integration… We’re not too far away from seeing a million-server data center.” This prediction by Kirk Skaugen, vice president and co-general manager at Intel Corporation, was part of an experts’ panel at the recent Windows 2008 launch, hosted by Al Gillen, vice president at IDC. (Archived video available here — requires Microsoft SilverLight download.)

While the vast majority of enterprises, of course, will not be faced with managing a million servers anytime soon, it’s fair to say that server proliferation and sprawl is now a pervasive problem, compounded by enormous growth in data and storage requirements.

Add to this the impetus to better align with the business, by being more flexible and adaptable to fast-changing processes. Given the prominence of the panel at the Windows 2008 event, this signals that data center efficiency is clearly an emerging positioning strategy for Microsoft – and perhaps a defensive one as well. Other leading vendors, including IBM and Sun, have been promoting this concept for several years, and have gained a lot of traction by blaming Windows-based servers for a lot of the server sprawl and waste that now dots enterprise landscapes.

Skaugen made the direct shot over the bows of the big server companies by pointing out that while distributed systems and storage put their share of strain on data center environments, large centralized systems are not above blame, either. “Intel and AMD today have 94 percent of the units of servers, but six percent of what is still RISC and mainframe computing make up almost 50 percent of the IT spend on hardware spend,” he said.

Regardless of platform environment, panelists agreed that the emphasis of data center value is rapidly evolving from “watts and slots” to achieving higher efficiency and better business delivery. The bottom line is that data center infrastructures need to deliver more for the business, while reducing their own footprints. “If you look at the emergence of these huge data centers, the new metric is going to be performance per watt per cubic foot – how much computational ability can you put into a volume space,” said Randy Allen, corporate vice president of AMD’s Server and Workstation Division.

James Mouton, vice president of the platform division within the HP Industry Standard Server business unit, suggested that executives and managers view data center metrics over at least a three-year time horizon. “You’ve got to look at that cost at least over a three-year period, including the power and cooling prices,” he said. “The manageability of the infrastructure overall is key. Unless you can see what’s going on, and map that to the application service level, you’re missing the forward-looking thinking. You have to make sure you’re looking at that very holistically, including business outcomes. It may or may not be a server uptime question — more likely, it’s a question of, ‘are your apps doing what they’re expected to do? Are your IT costs going down, as well as your overall three-year horizon of overall costs?‘”

Panelists also agreed that virtualization was clearly the best strategy for managing the complexity of server sprawl. Allen noted that the ability to pack greater processing power into smaller, more efficient spaces is a natural opportunity for virtualizing resources. “Virtualization flows right into that, offering simplicity in terms of managing all the complexity,” he said.

As enterprises demand more and more computing power, the challenge is more efficiently managing large-scale resources, and virtualization is one key element of such an approach. The challenges not only include improving the efficiency of data centers, but also reigning in energy consumption while correcting the underutilization of servers. As Skaugen put it, “The challenge that everyone’s facing today is that servers are only used 10 to 15 percent of their utilization in a standard data center, which is appalling.”

Rick Becker, vice president of software and solutions for Dell, added an additional element that should be considered alongside virtualization – intelligent network storage. “Intelligent storage is the next piece of the puzzle,” he said. “The only thing growing faster than server proliferation is data proliferation. As we go digital, we have to get intelligent about how we manage and grow storage. You also get more benefit out of a virtual environment when in fact the workloads are on a shared resource.”

Enterprises need to view their data center costs and benefits more holistically in terms of the business, panelists agreed. “In the old days, the question was on the entry price of the gear, and sometimes, enterprises would miss the whole, ‘once I deploy it, how do I deploy it, how do I maintain it what type of staff do I need, what type of training do they need to have?‘” said James Mouton, vice president of the platform division within the HP Industry Standard Server business unit.

Ultimately, the data center needs to view itself as a business, and not as a technology shop – a view echoed by another panelist, Ajei Gopal, senior vice president and general manager of the Enterprise Systems Management business unit at CA. Gopal said there’s an evolution underway, from “IT as a cost center to IT as a driver of the business.” Data centers need to think of themselves more as service centers, he added. As part of this initiative, Gopal advices data center executives and managers to embrace best practices such as ITIL [Information Technology Infrastructure Library].

Ultimately, through strategies such as adopting best practices and virtualization, many companies may be able to forego the need to dramatically expand or build new data centers to alleviate the crush of data and applications. “We have the ability to find the hidden data centers we have in our data centers,” said Becker.