Interview with the Expert:

Solutions for Data Center Management

Stanford University consulting professor, scientist, author and energy-efficiency expert Jonathan Koomey led the team that did the first peer-reviewed measurements of data center energy use, and he continues to lead efforts to improve energy efficiency in data center facilities. His 2008 study, Worldwide Electricity Used in Data Centers, described historical growth in total data center power use and gave urgency to industry efforts to improve data center efficiency.

Before offering The DCIM Advisory any solutions to the challenges facing data centers today, Koomey defines the four main business problems.

1. Underutilization of equipment leads to high cost per transaction. Koomey says, “Most servers in typical business data centers are utilized at only 5 to 10 percent of their maximum capacity, and cooling and power distribution systems are also used to much less than their full potential.  Clearly, that’s wasted capital, and it makes the cost per computing transaction much higher than it needs to be. A secondary effect is that the fixed energy costs for running servers at low utilization makes the cost per transaction much bigger than it needs to be.”
2. IT kilowatt-hours plus facilities support equal high overhead. “For every kilowatt-hour of IT electricity use,” Koomey says, “there’s another kilowatt-hour for supporting equipment — cooling, fans, power distribution and air flow. This is overhead that doesn’t result in more computing, and the more we can reduce it the better. We’ve seen some advanced facilities that use only 0.2 kilowatt hours per kilowatt-hour of IT load so there’s a long way to go for facilities designed in the typical way.”
3. The power consumption trend is high but slowing. While Koomey’s previous research confirmed that total electricity used by data centers doubled from 2000 to 2005, his current rough calculations show that the process is slowing due to the recent economic downturn. In addition, energy use has been reduced by improvements in the efficiency of equipment, optimized operations and virtualization (which drives up equipment utilization and reduces energy use and costs per transaction).
4. Impending legislation pressures organizations to lower power usage effectiveness (PUE). Koomey says, “Climate change is a serious problem that will eventually result in a tax on carbon usage. Recently, I did calculations for a data center with 130,000 square feet of electrically active floor area.  At about $20 per ton of CO₂ that facility would incur a $5 million cost per year for such a carbon charge in a region that uses coal-fired electricity, which is real money by any measure. Those who locate data centers in predominantly coal-fired electricity systems are taking on a serious business risk that most people now ignore, but they do so at their own peril.”

Rationalizing management decisions

Koomey says, “Most data centers suffer from responsibility and authority for the data center being assigned to different parts of the organization. For example, the facilities department has a budget for electricity and the capital costs of infrastructure; the IT department has a budget to buy computer equipment; the real estate department has a budget for building and landscaping. These separate budgets may lead to a problem if one part of the organization purchases equipment that forces the other part to spend money on energy or buy equipment. It’s the problem of split incentives: The IT department doesn’t care how much power its servers use and may not want to spend $1 on server efficiency, even though spending that dollar may save $5 for the facilities folks.”

When departments do not operate on a single budget, different departments, such as IT, don’t own the effects of the changes they impose. “A lack of coordination,” says Koomey, “borne of not bearing the effects of actions, leads to bad behavior.” The cost of this is no longer so small. In fact, a fraction of the total cost of owning and maintaining facility infrastructure turns out to be growing over time, as IT equipment gets more power intensive. “Nowadays, the implications of IT making independent decisions are more damaging to the parent company; 15 years ago, IT was the bulk of data center costs, but that’s no longer the case.”

Many data center managers don’t know what they’ve got on the floor. They don’t know how many servers are “comatose” or even how many there are, and it’s critical for good business practices to know that. Koomey says, “You not only have to bring it all together under one person and one budget, you need a tool that helps you see what you’re managing.”

Toward this end, there are many reasons to use data center infrastructure management (DCIM). Here are the top three:

• Visible data. Having visible data to manage increases your uptime and improves airflow; all this leads to more computing cycles. For this to happen, you must understand your inventory, power use, temperature and loads on the equipment.
• Efficient, flexible operations. Having the knowledge that DCIM offers allows you to act with more flexibility, speed and intelligence in the face of changes.
• Reduced costs. The promise of DCIM is that it will reduce costs if people use it appropriately. That means managers must have a clear plan, get staff on board and make use of the data.

If a company does not rationalize its management and institutional incentives, doesn’t have one budget for the facility and doesn’t develop a single, simple model for total cost of ownership, then the effect of doing all this measurement will be far less than if businesses take these issues seriously and actually implement management changes. In well-run data centers, critical environment teams have all the key players in the room when making decisions.

Koomey says, “Implementing DCIM is not enough. You have to change the way you use that information to make decisions. If you do that, then it can have a big effect, but you can’t just install new technology and expect it to save money.”

Learning from the PC revolution

To use DCIM effectively, companies need to take advantage of all it offers, instead of doing what they did back in the ’80s. People started buying lots of personal computers for companies; for years, the economic literature published many papers exploring the “productivity paradox.” People were spending money on computing, but the savings were not showing up in the productivity statistics.

Koomey explains, “People were plopping a PC down on someone’s desk without changing the way the institutions used information or made decisions. It took 10 or 15 years before companies started catching on to this and restructured their operations; finally, you started to see a massive improvement in productivity.”

It’s exactly the same story for DCIM. If you just install it without changing the way the organization operates, you will not see dramatic savings. DCIM is a key part of implementing change management.

Jonathan Koomey is a consulting professor at Stanford University, worked for more than two decades at Lawrence Berkeley National Laboratory and has been a visiting professor at Yale University (Fall  2009) and Stanford University (2004-5 and Fall 2008).  Dr. Koomey holds M.S. and Ph.D. degrees from the Energy and Resources Group at UC Berkeley, and an A.B. in History of Science from Harvard University.  He is the author or coauthor of eight books and more than 150 articles and reports and is one of the leading international experts on the economics of reducing greenhouse gas emissions and the effects of information technology on resource use.  His latest solo book is the 2nd edition of Turning Numbers into Knowledge:  Mastering the Art of Problem Solving. He’s also a coauthor, with John Stanley of the Uptime Institute, of the 2009 paper titled The Science of Measurement: Improving Data Center Performance with Continuous Monitoring and Measurement of Site Infrastructure, Oakland, CA: Analytics Press, October 23.

Be Sociable, Share!

•