A challenge for Google’s Deep Mind: solving the Jevons Paradox

Note: this post was originally published on The Fifth Estate on 16 August 2016. 

Google has announced that it has used Deep Mind, the neural network computing developed by its AI research company, to reduce the energy used for cooling its data centres by 40 per cent. Sounds impressive, until you realise that Google’s energy use is doubling every year.

Data centres are not very efficient for several reasons:

1. Servers are generally inefficient, producing lots of heat
2. Just cooling them uses a huge amount of power
3. The companies running data centres are rewarded for responding quickly to demands on server uses. They are not rewarded for saving energy, therefore they keep energy use at a maximum continuously
4. With the inexorable expansion of “cloud computing”, this trend is set to continue

So while Google’s announcement sounds great let’s look at what is actually happening to data centre energy use – in fact energy use in ICT generally. And it’s not good.

Even if data centres did use energy efficiently their overall use is still increasing at a greater rate. Google’s total power usage appears to have increased by a factor of 12 in the last four years, almost doubling every year.

It likes to boast about its use of renewable energy. It recently purchased 781 megawatts of solar and wind power to power its data centres. But the company also says renewable energy makes up just 37 per cent of its usage and with a total of 1.2 gigawatts of renewable energy, that makes its total data centre usage around 3.2GW.

But this itself is in the context of Google’s overall energy usage this year, estimated to be 48.927GW [for source see comments to the article linked to above].

The worldwide explosion of data centres and their increasing energy usage is a direct result of the spread of smart phones, tablets, apps and video-on-demand. We expect all of these things fast and free and that is what is fuelling the expansion.

The number of annually produced smartphones is expected to rise between 2010 and 2030 from around 350 million to around 3000 million units, and for tablets from 50 to 560 million units.

A worst-case projection for the global use of energy in ICT puts it at as much as 51 per cent of global electricity and 23 per cent of the globally released greenhouse gas emissions in 2030. Totally unsustainable? Right.

The Jevons Paradox

We seem to be seeing another example of the Jevons Paradox, the conundrum proposed by economist William Stanley Jevons. First postulated in the 1860s, it states that increases in efficiency will not result in savings, instead they will result in more expenditure or consumption.

Jevons argued, in his prescient 1865 book The Coal Question, that the effective and efficient use of energy leads to an increase in energy consumption. In his words:

“It is a confusion of ideas to suppose that the economical use of fuel is equivalent to diminished consumption. The very contrary is the truth…”

Increased energy efficiency tends to increase energy consumption by two means. First, increased energy efficiency makes the use of energy relatively cheaper, thus encouraging increased use (the direct rebound effect). Second, increased energy efficiency leads to increased economic growth, which pulls up energy use for the whole economy (the indirect effect). It applies to energy efficiency and resource efficiency.

When people save money on saving energy they have extra capital. This inevitably gets spent resulting in more consumption. The same applies to saving money on manufacturing products by reducing the amount of resources needed. More products get produced. Economists and environmentalists often use the amount of spending as a proxy for energy consumption or ecological footprint. The only way to really reduce environmental impacts is to spend, or consume, less.

As a whole, we on our beautiful unique planet Earth are already using two planets’ worth of resources and our numbers are rising, expected to reach 10 billion by the end of the century. This includes a growing middle class that is consuming more and more, causing some minerals and other resources expected to run out according to their relative rarity over the next century.

You can see why this is a vitally important problem to solve. It is a subset of the problem: how do we give everybody an equally of good standard of life given limited resources on the planet?

A question of entropy

Ultimately it is a question of entropy. For example, urban living can be seen as an entropy accelerator: resources are depleted and downgraded and the limited availability of low entropy energy is their ultimate constraint and a constraint on long-term well-being. Local air pollution and global climate change are a high entropy expression of burning fossil fuels to power lifestyles.

Another example is the problem of entropy of recycling materials. When recycled, most wastes come out of the end of the process as a lower-level product. Take electronic waste as one example, or paper as another (during recycling the fibres become smaller and so the paper is more fragile).

Circular systems, as used in nature, where there is no waste, need to be devised. If nature provides an example where entropy does not increase, can we apply this to industrial processes and human practices?

Now that would really be something for the Google geniuses who devised the algorithms for Deep Mind and learning artificial intelligence.

Mustafa Suleyman, the co-founder of Deep Mind, is already on record as saying that it may be possible to apply their algorithms to other scenarios.

“There’s lots of other applications outside of Google”.

Really? Life outside Google? You surprise me.

Further reading:

A simple introduction to the subject of entropy, pollution, the economy and human survival is found in this article: Energy consumption and entropy release in the biosphere.

For the brave, a more complex mathematical introduction is available in this academic paper: The Impact of Entropy Production and Emission Mitigation on Economic Growth.

David Thorpe is the author of:

• Best Practices and Case Studies for Industrial Energy Efficiency Improvement (with Oung, K. and Fawkes, S. UNEP, 2016)
• A London Conversation: Business Briefing on Green Bonds (The Fifth Estate, 2015)
• The One Planet Life (Introduction: Jane Davidson. Routledge, 2015)
• Earthscan Expert Guide to Energy Management in Buildings (Earthscan, 2013)
• Earthscan Expert Guide to Energy Management in Industry (Earthscan, 2013)
• Earthscan Expert Guide to Solar Technology (Earthscan, 2011)
• Earthscan Expert Guide to Sustainable Home Refurbishment (Earthscan, 2010)

Apple to power its cloud with the sun

Computer giant Apple has announced it will power its main US data centre entirely from solar energy by the end of the year.

It plans to invest an unspecified amount in solar generation capacity sourced from SunPower Corp, plus solid oxide fuel cell technology from Bloom Energy.

The fuel cells will supply stored power generated from the sun, when it isn't shining, and use a ceramic powder instead of platinum to produce electricity with greater efficiently than traditional fuel cells.

They operate at extremely high temperatures, typically above 800°C, which improves their electrical efficiency.

Bloom Energy already supplies Google, eBay and Walmart as well as Adobe Coca-Cola and other household names.

SunPower Corp was the foremost commercial solar installer in the United States last year.

According to Apple's CFO Peter Oppenheimer, the investment will benefit its North Carolina facility, which holds the servers that host its iCloud data services.

The solar farm will provide 84GWh of energy annually, more than sufficient to power the data centre.

This data centre is not the only one used for the iCloud by Apple, but it is the main one.

Greenpeace last month challenged Apple in its report “How Clean is Your Cloud?” that it was not as green as Facebook and Google.

Data centres contain thousands of computers that store and manage the world's rapidly growing accumulation of data for consumption at a moment’s notice. They consume a tremendous amount of electricity. IT in total is responsible for around 2% of global GHG emissions [this sentence corrected since publication - see comments below].

But Greenpeace says that most IT companies are rapidly expanding without considering how their choice of energy could affect climate change.

Greenpeace said that Yahoo and Google both continue to lead the sector in prioritising renewable energy to power their cloud expansion.

It accused Amazon, Apple and Microsoft of rapidly expanding without adequate regard to the source of electricity, and of relying heavily on dirty energy.

It applauded Facebook for committing to power its platform with renewable energy, and Akamai, responsible for carrying a much internet traffic, for being the first IT company to begin reporting its carbon intensity under the new Carbon Utilization Effectiveness (CUE) standard.

In response, two weeks ago, Microsoft promised that it would go carbon neutral after July. Its chief operating officer Kevin Turner said that it would use carbon offsetting and improved energy efficiency.

Greenpeace criticised Microsoft however, because carbon offsetting still allows it “to keep building data centres that rely on coal, such as its new investments in Virginia and Wyoming".

Apple, however, appear to have listened to Greenpeace's advice.

UK businesses are no more energy aware in their PC use than five years ago

UK businesses are wasting about £30.8 million a day by leaving their PC equipment on standby, according to new research from energy company E.ON – which, depressingly, establishes that nothing has changed in five years despite the huge amount of advice and support available.

E-ON, which itself offers energy efficiency advice, including online management tools for its business customers, says that more than one third of UK small businesses fail to power down computers at the close of day, and one in five only do so fully at the weekend.

It costs 1p per hour on average to keep a PC on standby, a seemingly insignificant sum, but one which can add up.

"10% of small business owners wrongly assume that leaving computers on standby at night uses the same amount of energy as shutting them down completely,” said Iain Walker, Head of Business Sales at E.ON.

“Introducing small changes into the culture of your business, like turning equipment off fully at night, can have a significant financial impact on your energy bills.”

The research suggests that there has not been much improvement in energy awareness in the last five years, since a similar survey was conducted by the National Energy Foundation.

This asked the same question of a representative sample of the general public's behaviour at work in 2006 and found more or less the same result: that as many as one-sixth of work computers are never switched off at night or weekends, with a further seventh not switched off on some days each week.

There is simple software available to automatically manage the power consumed by PCs or power down networked PCs centrally.

The NEF suggested that using this could save up to 1.5TWh (1.5 billion kWh) per year, with a value around £115 million, and 700,000 tonnes of CO2 (equivalent to almost 0.2MtC).

The question is – why has no one paid any attention to this information?

E.ON's research shows the retail sector includes the biggest culprits, with over two-thirds admitting that they do not switch off work computers at night. Even more worrying is that 59% only do so at the weekend, meaning that they stay on all week.

The farming sector contains the best performers, with 69% shutting down every night.

There are regional variations: small businesses in the North were found to waste the most money and energy, with 23% of small businesses in Yorkshire and 21% of small businesses in the North East believing that leaving computers on standby uses the same amount of energy as switching them off.

Those in the South were most energy aware, with only 4% from the South East, and 5% from London, believing the same thing.

It suggests that the Carbon Trust ought to target its work more in the worst-performing regions.

It also suggests that PCs should be shipped with a default setting that shuts them down automatically if not used for a couple of hours or so, which has to be disabled by the user - rather than the other way round, as it is at the moment.

As with the phasing out of incandescent lightbulbs, removing the choice to be inefficient from users without compromising the service they get, saves busy people from having to think about these things.

Efficiency help for businesses and data centre operators

In a related field, further new research from Analytics Press shows that the energy usage of data centres worldwide is increasing, but the rate of increase is slowing - at least up to 2010.

The total global electricity use by data centres in 2010 was about 1.3% of all electricity use for the world, and 2% of that for the US.

It increased by about 56% from 2005 to 2010 (instead of doubling as it did from 2000 to 2005), while in the US it increased by about 36% instead of doubling.

While Google is a high profile user of computer servers, less than 1% of electricity used by data centres worldwide was attributable to that company’s data centre operations.

Part of the eason for the slowdown in the rate of increase may be that there is now an increasing number of green data centres that are powered by renewable electricity or employ efficiency measures.

One pioneer in this area is Interxion, whose engineer, Ali Moinuddin, is co-chair of The Green Grid’s European Communications Committee and helps to spread awareness of advanced energy efficiency in data centres and the broader impact of business computing amongst the European community.

The Green Grid is a nonprofit network aiming to improve the resource efficiency of data centres and the general business computing ecosystem.

Many of Interxion's data centres now use 100% renewable energy as well as free cooling as standard and have cold aisle containment as a mandatory configuration.

Commenting on the research, he offered the following advice, not just for data centre managers but for all companies: "The Green Grid has developed the Data Center Maturity Model (DCMM), which provides clear goals and direction for improving energy efficiency and sustainability across all aspects of the data centre as well as clear metrics and guidance on what all organisations need to address within the whole ICT infrastructure in order to be more sustainable".

It allows users to benchmark their current performance, determine their levels of accomplishment, and identify the ongoing steps and innovations necessary to achieve greater energy efficiency and sustainability.