We present the important questions and answers that came up during the Data Center Power and Cooling presentation at the Gartner Data Center Summit held in London, 5 to 6 October 2009.

These issues will continue to plague users in 2010, and in this research we suggest pragmatic ways in which to deal with them. Among the prevalent issues are that:

• Energy costs are the fastest-rising cost element in the data center cost portfolio.
• Removing a single x86 server from a data center will result in savings of more than $400 a year in energy costs alone.
• Hardware assets should be used well after their depreciated financial life has ended — for as long as is technically possible.
• Data center and IT managers are not paying sufficient attention to the process of measuring, monitoring and modeling energy use in data centers.

We recommend that data center managers and IT managers:

• Implement improvements to your facilities — for example, raise the temperature of the data center and use energy management software.
• Implement virtualization quickly, and plan to have at least 50% of x86 data center servers virtualized by 2011 to improve asset use and to lower energy costs.
• Gradually raise the temperature at the server inlet point to run up to 24 degrees Celsius (75 degrees Fahrenheit), but use sensors to monitor "hot spots" and supplementary cooling, if required.
• Develop a dashboard of data center energy-efficient metrics that provide appropriate data to different levels of IT and financial management.

There is no single, standardized method to account for data center costs. Users need to define a chart of accounts that specifies all the cost elements that constitute the overall cost, and the key portfolios or categories (such as servers, software and networking) that are part of that cost.

Based on an aggregation of the actual end-user information that Gartner has collected regarding data centers, Figure 1 provides the composition of cost elements typically found in a data center portfolio. It also illustrates a typical percentage cost breakdown for a large (10,000 square feet or larger) data center for an end-user company, in an established market and geography (such as Europe, North America or Australia), that provides internal IT services (see "Managing the Data Center Cost Portfolio During a Recession"). We cross-referenced the figures with Gartner consulting benchmarking data, and we have excluded data centers for service providers and hosting companies, because the cost portfolio for these sectors will be considerably different. The actual cost level will depend on the location, industry and manner in which the data center is run; however, in general, the composition of the costs will be consistent among this category of users.

There are a number of pragmatic suggestions that are listed below (see "How to Cut Your Data Center Costs" and "Extend the Life of Your Server, but Beware Some of the Pitfalls").

Rationalizing the hardware involves taking out those systems that are underutilized, are old or where the workload can be run on more-efficient hardware. This will help with asset and inventory management, and will provide a clear picture of which hardware is being used effectively and which is not. Clients have reported that rationalization and consolidation programs have resulted in 5% to 20% fewer servers being deployed. Server rationalization should lower maintenance and support charges. These are normally around 10% of the discounted price and are based on the number of servers deployed. Reducing the number of servers will reduce the support costs.

Server rationalization will also lower energy costs. This is because less electricity will be needed to power the servers and to cool them down. A typical two-socket x86 server uses approximately 400 watts of power. At six cents a kWh, this equals $210 a year. An additional $210 will be used for cooling. Hence, removing a single server will save $410 per year.

Most organizations still have multiple data centers for their IT operations (see "2008 U.S. Data Center Conference: Data Center Build/Expansion Update"). In some cases, these are large, complex installations; however, there are often multiple small sites (less that 2,000 square feet) or even small machine rooms. Consolidating these multiple sites into a smaller number of larger sites will often result in financial savings.

Energy costs are rising for most data centers. Companies should employ tools and techniques to manage the energy cost curve:

• Raise the temperature of the data center to around 24 degrees Celsius, while ensuring that all equipment is certified at the new temperature. This will reduce the level of cooling required and, thus, the energy bill.
• Where possible, use outside/free air as an alternative to expensive air conditioning.
• Use hot-aisle/cold-aisle configurations, blanking panels and economizers.
• Use server-based energy management software tools to run workloads in the most energy-efficient way. This may include taking advantage of lower energy tariffs at different times.

People costs still form the single-largest cost element for most data centers, with users reporting that they can be as high as 40% of the overall cost. Examine the number of people needed to run your data center services and the types of skills needed for the next 24 months, with a view to reducing costs. In particular, benchmark the staff relative to the size of the data center and the type of services offered. An alternative method for managing people costs is to use managed service providers, but the service levels must be well-defined and negotiated.

Delaying the procurement of new assets is a necessary step for all data center managers. A server's useful life often exceeds its amortized life. Thus, upgrading a server based purely on the book value could incur unnecessary costs earlier in the life cycle. A typical x86 server is depreciated over three to four years, but can easily provide useful work for more than six years. "Extend the Life of Your Server, but Beware Some of the Pitfalls" shows that extending the life of an x86 server beyond three to five years can yield savings up to 32% on the average five-year spending for that server. Moreover, delaying the purchase of a new x86 server will result in price/performance benefits averaging 10% during a four-year time frame.

Energy management (both capacity and cost) can only be effective through advanced monitoring, modeling and measuring techniques and processes. "Green Data Centers: Guidance on Using Energy Efficiency Metrics and Tools" explains why this is so critical, and provides a snapshot of what is available. What is most alarming is that a Gartner interactive poll conducted in April 2009 shows that 48% of the 132 respondents have not considered the issue of metrics (see Figure 2 and "Data Centers Focus on Green IT, but Many Neglect Metrics").

Without metrics, it is impossible to get accurate data, which is essential to evaluating basic costs (how much energy is actually being used in a data center), proportioning those costs to different users and setting policies for improvement. These metrics form the bedrock for internal cost and efficiency programs. This is a new and complex area that needs to be examined very carefully, otherwise inappropriate data will be measured.

In summary, companies are not focusing sufficient attention on monitoring, measurement and modeling tools. We urge data center managers and IT organizations to make this area a high priority, which will be essential for the adoption of so many new technologies and adherence to government policies.

A Gartner poll (conducted at the Data Center Summit in December 2008 in Las Vegas) showed that more than 60% of respondents indicated that they would renovate or upgrade their facilities. Although respondents may not represent a statistically significant distribution (see "Power and Cooling Remain the Top Data Center Infrastructure Issues"), the results are consistent with Gartner client inquiries, which focus on whether refurbishing a current data center would represent better value for the money and be an easier-to-manage solution than building a new site or using a third party's facility. In addressing this question, the following points should be considered (see "Q&A: Beware of Data Center Refurbishing Projects, They May Be More Costly Than You Think").

Many businesses have data centers that have been acquired over many years through business growth. Not all of these data centers are ideally located. For example, some are in expensive cities where labor costs are high. Clients consistently report that in Tier 1 cities, such as London, New York City and Sydney, labor rates are up to 20% higher than in Tier 2 cities in the same countries. In contrast, the labor rates in emerging countries, such as Poland, may be more than 50% less expensive.

Companies need to:

• Evaluate the location of the data center in terms of labor rates, the cost of energy and facilities, such as transportation and telecommunications links.
• Carry out a political risk assessment of the location, focusing on security and the risk of terrorism.

Modern data center designs differ from legacy designs, and involve multimodular and multitiered build outs. However, the useful working life of a typical module tends to be between five and 10 years (see "Multitier Data Centers: A Hybrid Approach With Significant Cost Savings"). The essential consideration is whether or not the site is large enough to accommodate growth, given the investment required to refurbish and the long-term (10 or more years) scenarios the organization has for data center service provisioning. Ensure that the refurbished site provides at least five years of capacity (physical, electrical and networking) to make the project worthwhile.

Do not underestimate the amount of structural work needed to renovate a data center. In many locations, planning permission from local government offices may be necessary. To get approval, planners may need to introduce new, greener technologies, such as alternative and renewable energy sources. New facility components are expensive, but the real problem is in integrating new products within an existing building, and with existing components. For example, digital, rotary uninterruptible power supplies (UPSs) use different technologies than battery-based UPSs, and they need their own unique housing. Moreover, new building management systems, cooling technologies and cabling will all need to be integrated and engineered into an existing building, which is an expensive and complex endeavor.

Companies need to:

• Focus on the technical problems of integrating new facility components into an existing building. If this is too complex, then renovation may not be possible.
• Evaluate whether it's possible to renovate while still providing data center services. If not, then renovation may not work.