Quantifying data center performance
Kourosh Nemati, Aitor Zabalegui, Maira Bana, Mark J. Seymour · 2018
Power Usage Effectiveness (PUE™) has been proposed by The Green Grid (TGG) about a decade ago and it became a popular indicator of data center energy efficiency performance. Although, the definition of PUE is simple and straight forward, it can be misleading because it only considers the effectiveness of energy delivery rather that the resulting cooling performance or IT productivity that results. In addition, the measurement has fundamental limitations such as the inclusion of IT fan power in IT equipment power rather than cooling power and so the results can be skewed. Representing the thermal performance of a data center by a single PUE number is incomplete and a more holistic metric is required. The Green Grid (TGG) has developed a new view of multiple metrics (TGG Performance Indicator (PI)) that beside energy efficiency evaluate facility cooling performance based on IT thermal standards during normal and redundant cooling system failure. TGG PI defines the thermal conformance and resilience of IT as measures of cooling effectiveness along with PUE as a measure of cooling efficiency and displays them in a single view. This view allows the business, not just the technical teams, to observe the trade-off between adjusting thermal performance based on standards and striving energy efficiency and choose the balance that best fits the business need. The data center community is becoming increasingly aware of the need to provide efficient and effective data center. However, beside the publicized PUE figures for high profile hyperscale data centers what cooling performance can a data center operator realistically expect? The introduction of TGG PI offers a view to evaluate not just energy efficiency but the effectiveness of cooling delivery as well. However, does the view of the metrics tell us whether the data center performance is good or bad? This paper shows the PI values for a number of existing data centers and discusses the need for performance data to understand what performance might be expected for different data center types, cooling infrastructures, locations and associated climates.