Figure 1. Power in, heat out
Data center electrical efficiency is rarely planned or managed and, as a result, most data centers waste substantial amounts of electricity. Efficiency varies widely across similar data centers, and -- even more significant - the actual efficiencies of real installations are well below the practical achievable best-in-class values. In order to address this issue, users need a common metric in order to compare alternative data center efficiency scenarios. This will allow specific data centers to be benchmarked against similar facilities and to measure progress against internal or external targets.
The Metric
One of the common metrics in use is power usage effectiveness (PUE), which is determined by dividing the total amount of power entering a data center by the amount of power that actually makes it to the data center computer infrastructure. PUE is therefore expressed as a ratio, with overall efficiency improving as the quotient decreases toward 1. Data center infrastructure efficiency (DCiE) is another way to measure infrastructure efficiency and is the reciprocal of PUE and is expressed as a percentage that improves as it approaches 100 percent. Therefore, a data center with a PUE of 1.6 would have a DCiE of 60 (60 percent efficient).Figure 1 shows where power flows in a sample data center. Note that virtually all the electrical power feeding the data center ultimately ends up as heat.
The data center represented in figure 1 is a 50 percent loaded data center, with an “N” configuration for all systems (no redundancy), a traditional UPS (as opposed to a new high efficiency UPS), no economizer, perimeter cooling, poor floor layout, poor tile placement. Note that less than half the electrical power feeding this data center actually is delivered to the IT loads. The data center of figure 1 is said to have a PUE of 2.1 (47 percent efficient).
The energy consumption of a data center over a period of time is computed using the average of the data center efficiency over that period. Therefore, the importance of data center infrastructure efficiency is really as a tool to examine the average efficiency over a period of time. Single measurements of data center efficiency are inherently inaccurate and cannot be used as a basis for benchmarking or efficiency management.
Factors
Conditions in a data center change over time so the efficiency of the data center also changes over time. Several factors have a major impact on the data center’s efficiency, including IT load, outdoor conditions, or user configurations.
Figure 2: PUE as a function of IT load
Outdoor conditions also vary with time and affect data center efficiency. While various factors such as sunlight, humidity, and wind speed can affect efficiency, the most important variable is the outdoor temperature. The efficiency of a typical data center declines as temperature increases. This is because heat rejection systems consume more power when processing the data center heat and because outdoor heat infiltration into the data center becomes an additional heat load that must be processed.
Users take a variety of actions that affect the PUE. Users can change temperature or humidity set points, move or add vented floor tiles, or can fail to clean air filters. These effects are highly variable and depend on the exact design of the power and cooling systems.
When the user changes these settings by moving a vented floor tile or changing a filter or a temperature set point the data center design has been changed and new measurements are required.
A Mathematical Model
A mathematical model is the key to creating a process and system for efficiency management. It is the model that allows understanding of the causes of inefficiency; therefore, the purpose of data center efficiency measurement is to establish the parameters of the efficiency model.An efficiency model for a data center can be created for an existing data center, or it can be created before a data center is even constructed, if the design and the characteristics of the power, cooling, and lighting devices are known. If the model accurately represents the design, the data it provides will be similarly accurate. While the electrical performance of some types of devices, such as lighting, UPS, and transformers are very consistent and predictable, many uncertainties exist regarding the as-built performance of devices such as pumps and air conditioners that cause the model to lose accuracy. This is where measurement can help.
Periodic measurements can be part of an overall management strategy that includes both initial and ongoing measurements. These two types of measurements have differing objectives.
Initial measurements help calibrate the data center efficiency model, establish “as-is” and “should-be” performance, and identify potential efficiency improvement opportunities. Initial measurements typically require measurements on individual power and cooling subsystems, in addition to an overall efficiency measurement.
Ongoing measurements compare against the model to provide alerts of unexpected inefficiencies and to quantify improvements. Ongoing measurements can be made either by periodic sampling or by continuous instrumentation.
Data centers can be permanently instrumented for efficiency, or efficiency can be audited periodically using portable instrumentation. In either case, the power circuits to be measured must be identified. It is not necessary to measure the power flows in all of the thousands of circuits in a data center. Measuring the power flow in a small subset of the power circuits can provide very accurate computations of efficiency. The efficiency measurement strategy consists of the following elements:
- Deciding on permanent vs. periodic/portable measurements
- Identifying the appropriate measurement points
- Establishing a system for reporting efficiency data
- Instrumentation options
Periodic audits using portable instrumentation have a lower initial cost, and are particularly well suited for existing data centers near end-of-life. For new data centers, permanent instrumentation should be specified. Some power and cooling devices, such as UPS, may already have built-in power measurement capability. If the accuracy of this built-in capability is sufficient, built-in instrumentation saves the need to use any additional measurement instrumentation on the affected circuit. Efficiency management software should also be able to capture power measurements from power and cooling equipment with built-in power metering instrumentation.
The Model
Measurements are only useful when used in conjunction with models. For this reason, modeling is a critical aspect of efficiency management and the data inputs of the model establish the requirements for the measurement of power flows within the data center.Thousands of power flows are typical in an average data center. Analysis shows that it is not necessary to measure all of these flows in order to measure and manage efficiency. When combined with appropriate modeling and information about the power and cooling devices, it is possible to create an efficiency management system of high accuracy with only a small number of measurements.
Input 11 atwww.missioncriticalmagazine/instantproductinfo
