In a two-part blog on the subject of Power Usage Effectiveness, also known as PUE, Wai-Kit invokes the popular maxim, "what you can't measure, you can't manage," the truth of which is self-evident though the reality it describes apparently is not universally realized, according to the blog.

PUE was developed by the consortium The Green Grid. Wai-Kit describes PUE as "... a simple metric for data centers that is easy to understand and use, that enables DC operators and IT administrators to quickly estimate the energy efficiency of their data centers, compare the results against other data centers, and determine if any energy efficiency improvements can be made or if there are potential problems."

Problem is, the blog continues, the PUE metric – which is the ratio of a facility’s total power to the power being drawn by IT equipment – requires complete knowledge and understanding of each component in the data center and its power consumption.

Why? "Power being drawn by IT equipment" is a variable number, the blog explains. IT equipment's power consumption varies from time to time, depending on their utilization. For example, a server with 10% utilization will consume less power than a server utilized at 90% level.

The implications of this, in the blogger's view, are that the newest PUE standard published by The Green Grid outlines three levels of measuring data center energy, which addresses a continuous measurement rather than a point-in-time measurement:

• Basic (formerly called Level 1): Manual, infrequent collection and analysis of data.

• Intermediate (formerly called Level 2): Daily automated data collection and analysis, but at the facility level.

• Advanced (formerly called Level 3): Continuous, automated data collection and analysis at the device level.

In the second part of Wai-Kit's blog, the author notes that the most extreme measurement, i.e. Level 3, will enable trending of the efficiency curve of the data center facility over time. The second entry examines the relationship of PUE to the projected IT load or density, and the variance of energy consumption by facility infrastructure as IT load increases. The blog considers four scenarios: (i) 4kVA/rack IT load; (ii) 7.5kVA/rack IT load; (iii) 10kVA/rack IT load; and (iv) 20kVA/rack IT load. These scenarios assume the following:

• Modular design of 10 IT racks per module (POD) with maximum of 10 PODs or 100 IT racks in the facility; Use of closely coupled and localized cooling through in-row Emerson XDH units and Emerson Challenger ITR precision air-conditioning systems for each POD, and use of cold-aisle containment for each POD
• Modular UPS (160kVA per unit), N+1 configuration* (assumed power factor of 0.8)

• Tier-II classification

At 4kVA/rack or 3.2kW/rack of IT load, the blog continues, the total facility infrastructure load is calculated to be at 68.7% of the IT load, but as the IT load increases, this ratio decreases, to 42.3% at 7.5kVA/rack, 41.7% at 10kVA/rack and 39.6% at 20kVA/rack. It showed a significant drop in this ratio from 4kVA/rack to 7.5kVA/rack load, but beyond 7.5kVA/rack, even as the IT load per rack doubles, the ratio of facility infrastructure load over IT load did not significantly reduce further. This illustrates that there is a certain overhead of infrastructure components in data center facilities regardless of the IT load. Therefore, the lower the IT load, per unit of the IT load is going to have a bigger share of the infrastructure component load, the blog contends.

Wai-Kit's second observation is the growth of IT load need not necessarily incur a corresponding trend in infrastructure load. For instance, if the IT load is forecasted to be 7.5kVA/rack instead of 4kVA/rack, which means bumping up the IT load in each rack by 87.5%, the corresponding increase in infrastructure load is only a mere 15.34%. However, when the IT load increases beyond the 7.5kVA/rack, say by 2.5kVA/rack, the result is almost the equivalent percentage of increase in the infrastructure load. But as the IT load increases further, say from 10kVA/rack to 20kVA/rack (doubling it), there is a lower increase in infrastructure load, which indicates a curve where with low IT load/rack, infrastructure load high, but as the IT load initially increases from low density to 6kW/rack (or 7.5kVA/rack) range, the marginal return of each unit of infrastructure load is extremely high.

Moving across the curve further towards a 8kW/rack or 10kVA/rack range, it takes about the same amount of infrastructure load to support an equivalent additional unit of IT load, the blog asserts, and as one moves further across the curve into higher density IT loads, the infrastructure becomes slightly more efficient again and gradually increases its marginal return. A slanted V-shaped describes the marginal return of infrastructure load per IT load across different range of IT loads/rack.

Wai-Kit concludes that, if you take the above observation into consideration, the resulting PUE is that it is initially not very significantly attractive, at 1.687 for 4kVA/rack load, but when the IT load increases to 7.5kVA/rack, it improves by more than 15%, down to 1.423. However, the marginal improvement in PUE does not track the same trend as the growth of the IT loads/rack. Even at 20kVA/rack, the calculated PUE is only at 1.396.

Generally, Wai-Kit notes, data center facility infrastructure loads does not mirror a straight line correlation to IT loads. Moving from a very low density situation to 6kW/rack level, produces a very significant improvement in the PUE.

More Information

PUE, Part 1

PUE, Part 2

Maximize the Value of Your Datacenter

The Green Grid [...read more...]

		fullsource
Print Email Current Profile Archive Blog Generate newsletter