by Victor Avelar
Before data center infrastructure efficiency can be benchmarked using PUE or other metrics, there must be agreement on exactly what power consumptions constitute IT loads, what consumptions constitute physical infrastructure, and what loads should not be counted. Unfortunately, commonly published efficiency data is not computed using a standard methodology, and the same data center will have different efficiency ratings when different methodologies are applied. This paper explains the problem and describes a standardized method for classifying data center loads for efficiency calculations.
PUE is defined as the ratio of two numbers, data center input power over IT load power. While it at first appears to be a problem of simply obtaining two measurements and taking their ratio, it rarely is this simple in production data centers. Data centers are often part of a building that has multiple purposes, so it can be impossible to find either a single point to measure the total data center input power or to measure the IT load. This means that the appropriate measurement points to capture all data center energy use need to be identified, and typically the power use data from different subsystems needs to be combined to find the total data center input power. To further complicate the matter, some of these power measurements may be impractical to obtain, or the devices may be shared with
non-data center applications.
This paper has provided lists of the various data center subsystems that need to be included in energy use analysis, along with how the energy use of these subsystems should be incorporated into PUE calculations. In addition, the paper has described practical approaches to determining the energy use of shared devices and devices that are impractical to measure, in order to allow for PUE calculations. Data center operators need to understand that determining PUE does not require extensive, expensive instrumentation because many losses in a data center can be very effectively estimated by indirect measurement and estimation. While these methods appear difficult for typical users to implement, low cost, easy to use software tools are available to simplify the problem and allow any data center, large or small, new or retrofit, to have an effective real-time energy management system.