Dr. Albert Esser, vice president of power and data center Infrastructure solutions at Dell, discusses how changing your mindset can help meet compute demand for years to
come. By setting operational policies around virtualization, regular hardware refreshes, and best-practices data center design, IT leaders can help improve productivity and lower
power consumption—enabling much more useful work to be performed within facilities that are already in place.
Standard measures of data center efficiency focus exclusively on how a computing infrastructure uses the power flowing into it. Given that many data centers are reaching the limits of their power and cooling capabilities, these are important metrics. However, a second and equally important consideration can also affect the balance sheets: server utilization. To unlock the true potential of the data center, enterprises must shift their focus from power consumption patterns to the overall productivity of their IT environments. Operational policies designed to increase server utilization and advance overall performance and efficiency can lead to dramatic improvements in data center productivity without increasing power consumption. Adopting these policies enables administrators to support compute demands for growing business requirements within their existing data centers—instead of having to build a new capital-intensive facility to process the required workload. Dr. Albert Esser, vice president of power and data center infrastructure solutions at Dell, spoke with Dell Power Solutions recently about a new metric to measure overall data center effectiveness, why utilization
is so important to data center productivity, and how organizations may dramatically improve their data center productivity while still staying within the boundaries of limited power supplies.
What is an accurate gauge of data center effectiveness?
One way is to look at how efficiently power is consumed. The industry-standard metrics commonly used to measure data center effectiveness are Power Utilization Effectiveness (PUE) and Data Center Infrastructure Efficiency (DCiE). Both are useful metrics, but they can be misleading because they do not tell the whole story. Although PUE and DCiE effectively measure the efficiency of power utilization within a data center, they are not designed to capture the amount of actual productive work being completed. For example, a data center may be very power efficient—that is, have an excellent PUE rating—yet still not be operating
anywhere near its full compute potential. At Dell, we believe a useful measure of data center
effectiveness must include actual work completed per watt, not just power efficiency. We have proposed a new metric, data center performance per watt, which directly measures actual work completed relative to power consumption (see Figure 1). This type of measurement
captures not only power efficiency, but also the effectiveness of computing resources in doing actual work. A useful analogy is car mileage—no one is too concerned about how many gallons a gas tank holds; what people really care about is how many miles a car gets per gallon of gas. Same with a data center: for administrators who have hit the wall in terms of power availability, the pivotal question is, how far can their data center go on the same amount of power? Another useful metric for measuring data center productivity is data center IT utilization,
which measures actual work completed relative to data center compute potential
(see Figure 1). This measurement captures how effectively a data center is taking advantage of the compute power that is already in place. In practice, actual CPU and network utilization are very good indicators for IT gear in production; exact measures for what constitutes useful work are still being developed, but it is important to start thinking about data center productivity
in this way to help determine suitable operational policies. The proposed productivity measures scale from server to facility, and can be used to guide purchase decisions. Today, industry-standard load simulators are good estimators of actual performance in production. At Dell, we believe a holistic metric will help drive efficiency and productivity and enable data centers to compute more while consuming less—leading to a reduction in servers, space requirements, and power consumption.
By Dr. Albert Esser, Vice President of Power and Data Center Infrastructure solutions at Dell
Download Entire Article PDF