For the past decade, Power Usage Effectiveness has been the most common standard to measure data center energy efficiency. While PUE remains in the news with recent controversy over its inclusion in the latest ASHRAE standards, other energy efficiency metrics are starting to catch on – specifically server utilization.
We’ve covered PUE before on the blog, but basically it’s the ratio of overall power used to power used for strictly computing equipment. The closer to a 1.0 ratio, the more efficient the facility.
As the industry has matured, PUE has come under fire as being too simple, easy to manipulate, or failing to consider other environmental concerns. This led to the development of other data center energy efficiency and environmental impact measurements and benchmarks, for renewable use, reeuse of energy, and even water consumption.
PUE remains a useful metric as a baseline “at-a-glance” look at the efficiency of any given facility, but there are a lot of factors at play. For example, we studied the carbon consumption of data centers across the USA with the same PUE but located on different electric grids. The location of a data center means even if you have 1.2 PUE, you could emit more carbon than a 1.2 PUE facility located on a more efficient local grid.
ASHRAE, an organization that develops building code standards for cooling systems and electric consumption, has certifications and standards in place for data centers. Their latest draft data center standard was set to include PUE minimums. In order to address the disparate efficiencies of electric grids across the world, they attempted to use different geographic zones. A given facility might need a PUE of 1.3 – 1.6 depending on their local zone.
In March, ASHRAE dropped PUE from the document, largely due to concerns from colocation providers. Because colocation facilities remain unoccupied or partially filled for many months or even years, their PUE rating suffers, as they must cool, heat, move air, and light areas of the facility that do not house any computing equipment yet.
The latest draft instead includes a combination of MLC and ELC, or mechanical and electric load components. This should allow more flexibility, as a highly efficient electrical system could compensate for inefficient mechanical aspects like overcooling.
The Green Grid, creators of the PUE metric, is also looking at ELC and MLC for possible inclusion in their Data Center Maturity Model, to be released in 2017.
The Uptime Institute released a study in October 2015 that revealed the measurement of actual server utilization is becoming more popular in many data centers. This is especially important as so called “zombie servers” consume a dramatic amount of energy while providing zero or very little computing benefits. The Wall Street Journal reported in 2015 that globally, underused servers account for the equivalent electric production of eight large power plants.
The Uptime Institute survey found that PUE remained the top priority for data center metrics, while server utilization stood at fourth most important, behind operating expenses and kW per rack as well as PUE. In 2014, as a contrast, 45% reported that they did not conduct any auditing on a regular basis to find underused or idle servers in their data center.
Awareness of server utilization can only lead to more efficient facilities, better use of virtualization (which can enable servers to squeeze more computing power out of limited resources), and more streamlined IT operations.
The federal government has made a strong push to consolidate data centers, largely due to the expense and complexities involved in managing dozens of facilities that really aren’t doing much computing. Other organizations might be wise to follow their lead. Stamping out zombie servers saves electricity consumption, lowers carbon footprints, reduces operation expense, and improves PUE ratings.