The Road To a Smart Data Center
Historically, all data center owners/operators have benchmarked the efficiency of their data centers using power usage effectiveness (PUE) and data center infrastructure efficiency (DCiE). This approach has led to the well-known “PUE wars.” The Green Grid, author of both PUE and DCiE, didn’t intend for these metrics to be used for the purpose of comparing one data center with another. Unfortunately data center operators continue publishing PUE numbers in an attempt to market their facilities or design strategies. Although efforts to improve data center efficiency should be celebrated and publicized, it is important to understand that these metrics by themselves are inadequate in determining data center efficiency, they are only the beginning. The data center efficiency metrics must include “productivity.” Are you getting the most “bang for the buck” from your servers and storage? Are you maximizing processing power? Are you retiring idle servers? Are you consolidating and virtualizing? Are you using renewable energy as opposed to conventional energy sources? Are you using wasted energy in a productive manner? A new approach should be adopted to look beyond PUE and avoid missing out on the opportunity to manage other major aspects of power consumption if we truly want to compare one data center’s design and operations with those of another.
Several technical papers have provided estimates and actual measured PUE numbers for data centers. In its report to Congress, the U.S. EPA provided a set of efficiency-improvement scenarios that predicted future data center energy consumption and associated PUE values to 2011. The EPA listed four categories of data center efficiency improvements, with increasing cost and complexity, summarized in Table 1.
Total facility power comprises a number of data center systems. Typical power consumption for various systems is provided by The Green Grid data center model shown below.
Detailed Data Center Model
[data center model] The above data center model can be expanded to include other factors such as environment, use of natural energy, recovery of wasted energy and so on. In this model, the total load is divided into IT equipment such as servers, facility load such as air-conditioning/power-supply systems and operation for each type. Electricity is sorted into grid power and green power generated in the data center. New data center construction would use photovoltaic (PV) panels mounted on the roof, whereas existing data centers can use thin-film technology should their existing roof construction be unable to handle the stress and weight of PV panels and associated framework.
Developing the New Metric
The Green Grid has proposed a new metric called data center energy productivity (DCeP). Like PUE and DCiE, time is important to this measurement. DCeP has an assessment window where useful work and energy are compared relative to a user-defined time limit.
[data center energy productivity] The useful work produced has been proposed as the product of the sum of all tasks, the value of the tasks, a time-based utility function and the absolute time of completion. The mantra of data center gurus at the moment is right-sizing; calculating DCeP allows users to right-size virtual and physical infrastructures to support business needs.
This metric, however, is difficult to implement, especially for colocated data centers. Measuring the productivity from each lessee would be tedious and, in certain cases, impossible. DCeP excludes factors that take into account various energy-saving strategies and will thus be insufficient.
Expanding on the research done by the Environment Department at the Japan Information Technology Service Industry Association (JISA), we can define a better metric that not only includes majority of the energy saving strategies but also provides sub-metrics to gain an overall green data center title. These sub-metrics focus on personnel and equipment safety, data center efficiency, and sustainability or reliability.
Enterprise-Level Real-Time Power Management
A real-time power-management system not only calculates each of these metrics in real time, but it also provides dynamic dashboards that can be published on websites worldwide, highlighting data center optimization and energy-saving techniques.
Today’s power design and management software enables designers and engineers to conceptualize the power distribution model for mission-critical facilities, simulate and test the integrity and security of the system, and analyze the results with accurate reports. Modules, such as for an underground raceway system, could enable data center professionals to design and predict the system availability not just for today, but for future server-farm expansion.
One-Line Diagram: Design to Sustain
It’s no secret that data centers are power hogs. Too many are using more power to cool the facility than to operate the computers that live there. Creating a one-line diagram of the data center is the first step towards measuring and managing overhead power demands. A one-line diagram allows centralization of a system knowledge database, which includes parameters of the electrical components that will help determine the electrical behavior by creating different scenarios. Creating numerous combinations of networks having diverse configurations and varying engineering properties allows users to fully investigate and study the behavior and characteristics of the electrical networks using one database. A series of “what if” scenarios can be designed, allowing the engineer to predict any future behavior of the power system.
Advanced Monitoring: If You Can’t Measure It, You Can’t Manage It
[energy monitoring] The monitoring system should be configured to allow the operator to replay previously recorded message logs while controlling playback to rerun at original or accelerated speeds. The ability to recover from a system disturbance depends on the time it takes to establish the cause of the problem and take remedial action. This requires a fast and complete review and analysis of the sequence of events before the disturbance.
Power-management software should assist your operation and engineering staff to quickly identify the cause of operating problems and determine where energy costs can be reduced. Besides reducing losses and improving data-gathering capability, such an application should assist in increased plant reliability and controlled costs. An event-playback feature is especially useful for root cause and effect investigations, improvement of system operations, exploration of alternative actions, and replay of “what if” scenarios.
To obtain optimal results, create a trend of the monitored and archived data and predict future results. An estimate can be obtained using the nameplate data and results from the original design of the one-line diagram, but this approach will lead to overestimation because the nameplate data represents the power that can be produced, not the actual power required to supply the servers. With the mission-critical facility online, monitor the active power levels and energy consumption while tracking costs. Also, create a report that displays the PUE and DCiE, thus providing a method to determine operational efficiency and allowing you to view the effectiveness of new energy-saving projects.
a power-management solution can enable optimization of power usage and energy savings while maintaining high availability and reliability. Optimizing power usage can reduce total cost of ownership (TCO) and ensure a profitable life cycle for a mission-critical facility. Extending the power-monitoring system by equipping it with an appropriate electrical-system context, simulation modules and playback routines will give IT personnel a powerful new set of tools. Finally, all of these capabilities should be included in one application that gives you the flexibility and compatibility to expand and upgrade your power-management system as your needs change.
Leading article image courtesy of IntelFreePress under a Creative Commons license
About the Author
[Tanuj Khandelwal] Tanuj Khandelwal is Vice President of Engineering Services for ETAP. Mr. Khandelwal received his BS in electronics and telecommunications engineering from University of Bombay and his MS in electrial engineering from California State University, Long Beach.