Sustainable Information Technology

What is PUE?

The aim of this post is to explain the background to Power Usage Effectiveness (PUE), what it is, what it’s used for and what people are saying about it.

The concept of ‘Green IT’  is a fairly new one. As technology changes and in turn transforms society, our understanding of what Green IT is and should be develops alongside it.  But the basic idea of Green IT is to make sustainable information technology a reality in businesses and in the home. One of the ways we can do that is to use metrics which help us understand and measure how efficient and sustainable IT actually is. PUE (Power Usage Effectiveness) is just such a metric.

Social Networks and the Cloud and eCommerce and Mobile Computing and… and… and…

PUE Power Usage Effectiveness But before we get into the specifics of what PUE is, let’s think about how these changes in technology are playing out in the real world. The global economy and society is driven by IT – in the huge expansion of social networks, eCommerce applications delivered from the cloud and the many flavours of mobile computing. In the developed economies we tend to relate these developments to cool stuff – people, places, games and applications. We equate celebrity tweeting, Silicon Valley (or Silicon Roundabout), viral marketing of Hollywood blockbusters, the latest iPad release and so on with what the internet is all about, and in some ways we’re right.

But many commentators have noted how the use of mobile ‘phones to find and share information is transforming day-to-day life for the poor in developing countries. This is an example of how the demand for information and connectivity will continue to grow exponentially for some years yet (see the United Nations Broadband Commission for Digital Development for some interesting background on global connectivity).

What these technologies have in common is their need to deliver information from data centres, the often immense buildings which house the servers and other computers which hold the information that’s presented to us through the web – in our offices and homes and on our PCs, ‘phones and tablets.

Those servers and computers need power: lots of power.

There are also other critical pieces of IT equipment that handle connectivity – such as switches, routers and load balancers. In the PUE equation, all of these pieces of kit taken together are what is referred to as IT Equipment Power.

IT’s all about Data Centres

But let’s think about this and what a modern data centre is like: potentially thousands of servers; potentially hundreds or thousands of network devices – switches, routers and so on… and all of this in a confined space. The amount of energy (the IT Equipment Power referred to above) being consumed is immense – but so is the heat being generated. So one of the first jobs of the data centre is to keep all of this equipment at acceptable operating temperatures. Like any other building, there are also other demands on the power supply. So the Total Facility Power part of the PUE metric includes items such as

  • IT Equipment Power (as discussed above)
  • Computer Room Air Conditioning (CRAC) Units and other kinds of cooling systems
  • Lighting
  • Power Distribution Units (PDUs)
  • Distribution losses
  • Generators
  • Uninterrupted Power Supply modules (UPS)

At last: the PUE Metric

Now that we know what is meant by IT Equipment Power and Total Facility Power we can start to understand what the PUE metric is really getting at.

The thinking behind the Power usage effectiveness (PUE) metric is that the efficiency of the data centre can be indicated by how much total facility power is needed in proportion to the power consumed by the IT Equipment alone. So if we had a scenario where total facility power was twice as much as IT equipment power alone, we would be running an inefficient data centre. The ideal PUE value is therefore 1.0 – a figure that would indicate that the ONLY power being used was that consumed by the IT equipment itself. Another way of stating this is that as the usage of power becomes more effective, the total facility power value and IT equipment power value will move closer to being equal.

The PUE metric was developed by The Green Grid. The inverse of PUE is DCiE, or Data Center infrastructure Efficiency.

Criticisms and drawbacks of PUE

Several criticisms have been levelled at PUE. These criticisms have to be seen in the wider context of Green IT, which is why I started this article with a description of what Green IT is trying to achieve. Data centres themselves, though big and complex – don’t exist in a vacuum. For starters, they are connected to power grids, and this is the source of the first criticism of the PUE metric:

Gripe 1: PUE doesn’t care about where data centres get their power from

Some electricity is generated with low carbon emissions. Other sources of power are ‘dirty’. PUE doesn’t discriminate between them. This was articulated in the 2012 Greenpeace report How Clean is Your Cloud? This is what Greenpeace said:

Some companies (are) inappropriately using metrics such as Power Usage Effectiveness (PUE) in place of meaningful metrics that would speak to the actual performance of their data center(s) in terms of computing resource or the natural resource being consumed in generating electricity.

In its defence, PUE wasn’t designed to measure total carbon emissions. To do so, it would be difficult to know where to start and where to stop. There is undoubtedly a huge role for measuring the carbon efficiency of electricity generation in the move to Greener IT, but PUE shouldn’t really be held accountable for not doing something it wasn’t designed to do.

What Greenpeace are doing here is really less about the usefulness of PUE, and more about putting pressure on the big data centre owners and operators such as Google, Microsoft, Apple and Amazon, so that they move towards ‘clean power’ more quickly. I applaud them for doing so.

Gripe 2: PUE can be deceptive and misleading

This is the most serious charge aimed at PUE. Again the Greenpeace report points out that:

in some circumstances, (PUE) penalizes better performance. For example, if a… manager identified servers in their data center that were not being used, and
elected to shut them off and create virtual servers, as shown in the table below, this could result in decrease in the power consumption rate (good) but an increase in the facility’s PUE (bad).

This is a serious objection indeed. In the real world though, it’s likely that turning off the servers and getting that 5mw saving would also produce more than a 5mw reduction in the total data centre power demand, which is the whole point about the PUE measurement. In other words, the switched-off servers would no longer require cooling, could be moved out of the DC server to a holding area and thus potentially improve air flow in the server room itself, be removed from the monitoring tools (a minimal saving, true) and possibly some other things I haven’t thought of, all of which could further reduce. total power demand.

Gripe 3: There are better measures than PUE

Again, in the same report Greenpeace encourage the use of another metric developed by The Green Grid: CUE, or Carbon Usage Effectiveness. This provides a carbon per kilowatt hour intensity measurement and has not yet been widely adopted by the industry. It’s great that Greenpeace are puching for more intelligent measures.

Conclusion

Like any metric, PUE has to be used in the right context. It would be great to have holistic measures of everything, but that isn’t practical. Used in the right hands there can still be a role for PUE, at least in exposing the most glaring inefficiencies in data centre management.