Early this month, I attended the Technical Forum of the Green Grid vendor consortium. The
Green Grid is a recently formed group that brings together major businesses to establish useful tools and policies for eco-responsibility in IT shops. Its activities include defining metrics for the IT industry, establishing best practices, and encouraging adoption of both.
The two-day forum was narrowly focused on the quest for useful, usable metrics that measure energy efficiency in data centers. While many members of the technical committee have been working on this problem long before Green Grid existed, I was surprised by how little consensus there was on how to measure energy efficiency and how crude the proposed measures currently are. This observation does not in any way denigrate the efforts of the committees; rather it points out how terribly difficult it is to formulate useful indices.
Let me give you an example and you'll see the problem. Then I'll try to give some guidance on what you can do until the Green Grid, or some other organization, comes up with guidelines and measures we can all use.
The principal measure presented at the Forum was called data-center infrastructure efficiency (DCiE). This metric is the ratio of power consumption by IT equipment to the total power consumed by the data center, expressed as a percentage. Or, put algebraically:
IT power consumption is defined as the power consumed by true IT functions and devices: servers, networking gear, etc. The larger data center consumption includes all the IT consumption plus cooling, lighting, and other common power sinks.
This ratio posits that the more overall data center power driving IT equipment, the more efficient the data center is. In other words, a higher DCiE ration represents a more efficient data center. Before discussing all the limitations of this metric, let me give you a sense of the DCiE ratios that the Green Grid expects to reward:
At 50 percent, a site is at the lowest level that merits an accolade (the so-called "bronze" level). At 90 percent, a data center crosses the highest recognized level (platinum). This range presents a very crude gauge of where your data center should be. However, it's obvious that many, many adjustments have to enter into the DCiE computation before it can be useful.
The first, of course, is geography. Cooling a data center in Nome, Alaska will be require less power than the same data center in Houston, Texas properly chilled. The next problem is the kind of data center you're running: a Tier I center will have a very different profile than a Tier IV center. Green Grid proposes adding credits and penalties to DCiE to account for these factors. However, the formulas for these adjustments are not ready nor, it seems, close to release.
Another factor is that comparing DCiE ratios makes sense only within a specific industry sector. Surely, a financial center will have a very different efficiency profile than say, a chain of funeral homes. As a result, the DCiE will probably be assessed qualitatively by industry. Whether this changes the proposed formulation, or recasts its interpretation as relative to an industry benchmark, is hard to know.
A final point is that DCiE measures the efficiency of the data center, not of the IT equipment. In fact, replacing IT equipment with more power-efficient gear will lower, rather than raise, the DCiE, unless an even greater reduction of cooling power is implemented. Unfortunately, cooling power consumption and IT consumption don't move in lock step and so there will be gaps in the figures where sites are not rewarded for investment in energy-savings equipment. Surely, that has to be improved.
In this entire discussion, I have not touched on the difficulty of measuring IT power usage as well as the power consumed by the entire data center. Neither of these tasks is easy. The upshot is that the industry has a long way to go before useful metrics can be established for data center efficiency.
Despite this limitation, the 50 percent mark that the Green Grid proposes for recognition is a useful figure. It suggests a minimal level that is worth working towards. Doing so begins with figuring out how exactly you're going to gather the data. Once you have usage data that you feel is comfortably representative, you can begin monitoring the trend and figuring out what you can do to climb over this minimal threshold. After all, why is your data center infrastructure using more power than your IT equipment? If you don't know, this would be a good time to begin finding out.
