The Internet is Frying the Planet: The Issue of Massive Electricity Consumption from Information and Communications Technology and Data Centers

The Internet is Frying the Planet: The Issue of Massive Electricity Consumption from Information and Communications Technology and Data Centers

Our communications technologies combined with the data centers that back them up are immense users of electricity. In 2014 data centers in the U.S. were consuming about 86 terrawatt-hours (TWh) per year, which is about 2% of domestic electricity. Another report pegged it at about 70 TWh. Globally, the estimate is about 1.3% of global electricity. By another estimate of cradle-to-grave analysis of the total energy consumption of information and communications technology (ICT) the total came to about 1500 TWh/year, or 7% of global electricity. Some think that number omits quite a bit of data-dependent digital technology and the real number is even higher. Indeed, that number is undoubtedly higher in heavily technologically-equipped countries such as the U.S. ICT power usage breaks down into 1) end-user devices like PCs and smartphones; 2) wireless networks; 3) manufacturing; and 4) data centers. As of 2016 it has been estimated that electricity demand for data centers is set to double in the next five years. Data center energy usage growth actually leveled out after 2005 due to efficiency improvements but continues to grow at slower rates than previously. 

Another looming issue that is set to add to this energy demand is the IOT and smart devices revolution, including the software and controller devices that run renewable energy demand response systems. This will increase data usage significantly and thus should be a consideration in evaluations of overall emissions impacts. The use of data-intensive, software-controlled, real-time automation switching and attenuating is being applied to many industries for good reasons – increased efficiency, cost savings, and improved safety. However, the vastly increased data usage is a clear downside to these tech developments. 

Video is an especially big data user and its use is being promoted by Facebook, Youtube, and others, and smartphone photos and videos are moving to cloud storage. As much of the world moves to wireless information and communications technology (ICT), energy use will increase. Smart phones using broadband represents the biggest jump in traffic and this is proceeding much faster than data and energy efficiency increases. The vast energy needs of broadband wireless smart-phone enabled ICT are globally powered mostly by the world’s top energy source – coal. That is changing slowly but will still be the case for some time. Data is cheap so more people use more of it and energy use goes up. China and the U.S. are leaders in cloud computing and China has recently built a whopping 200 MW data center touting cheap energy (from coal).  

  

So-called “tech” companies like Facebook, Google, Apple, Microsoft, and Amazon are now some of the biggest and most profitable global companies. They also consume large amounts of energy. Google’s attempts to utilize renewable energy on a vast scale have generally failed due to scaling problems and the shear amounts of energy required for their data centers in populated areas where there is simply not enough space for the land footprint required for such renewable energy systems. Much of the energy is required for cooling. Promising research is being done to increase energy efficiency for data center cooling. Google recently announced results of a pilot project utilizing artificial intelligence (AI) that led to 40% less data center energy requirements. Such improvements will be necessary just to slow the growth of data center energy use. Similar improvements include algorithms that adapt to usage. Other improvements include “smart antennas, spectrum sharing, new power supplies, and amplifiers.” The addition of miniature cell towers and in-building Wi-Fi can help relieve data traffic congestion. While the Google study is good news it should be pointed out that the big tech companies like Google are only a small part of data centers. Many smaller and much less efficient data centers occupy space at universities, government, and midsize enterprise data centers house the bulk of the world’s IT equipment. They are plagued with inefficient cooling as well as overcooling, which is inefficient. Overcooling is typically due to poor air management systems. Cooling hot spots requires a certain amount of air but other spots don’t so there is much waste. This happens simply because reliability needs trump efficiency. Better automated controls can help. Better hot air/cold air separation is also a general need. Cooling can use up to 50% of data center energy so this is important.

After Greenpeace rated the big tech companies by their use of renewable energy in 2012 there was another estimate made the same year by James Hamilton an engineer at Amazon. He calculated how much land space would be required for solar panels to power a new Apple iCloud data center in North Carolina. He estimated that powering the 500,000-square-foot center with solar panels would require a whopping 6.5 square miles (16.8 square kilometers) of tree-less, building-less land, all in a populated area! Nearly 11 square miles (28 square kilometers) of wind turbines (with avg. power density of 1 watt/square meter) would be required to provide the 28 megawatts of power needed for Facebook’s data center in Prineville, Oregon, estimates Robert Bryce. A few data centers are nearly ten times this size! He also concludes quite convincingly that “Green” Computing Can’t Power the Cloud.” Another way Bryce puts it is that Big Data requires Big Electron. The daily cycle of wind power output is almost opposite the daily cycle of data traffic so wind is not a feasible energy source for data centers or other ICT for that matter. Solar output is more in line but the land availability problem nixes it for now. However, storing data on the cloud does indeed save both money for consumers and providers through sharing resources, so overall in most current cases the cloud saves compared to storing on devices. The problem with the cloud is that using it requires frequent high-speed accessing and transferring of data. 

A recent plan for a Facebook data center to be sited in either Utah or New Mexico plans to be powered mainly by solar (day) and wind (night) in those states with good resources of both and plenty of available land. Such a situation would not work in the more populated areas of the eastern U.S. There is a significant effort underway to supply data centers with renewable power. While smaller companies may offset their coal and gas use by purchasing renewable energy credits the bigger companies are building solar and wind farms. The big tech firms have been teaming up to push for more support of renewable energy, presumably to help offset their increased use of non-renewable energy. The big tech firms are leveraging their large power use (much of it now due to cloud storage) to get the utilities to add more renewables. This is acceptable and sensible in some ways, of course, but one also needs to consider that since renewable energy is directly subsidized by taxpayers, one could also argue that in a sense they are also pushing for more taxpayer subsidization of their energy consuming centers and using up government subsidy resources as some data centers require whole mid-sized power plants just to run them. Greenpeace noted the basic formula for greening IT = energy efficiency + renewable energy but that is no easy process and in many places and situations is simply not feasible at present. Of course, wind and solar need baseload capacity power, most often fossil fuel sourced, for reliability and back-up. Greenpeace also noted the usefulness of siting the centers where renewable energy is available and where there are favorable renewable energy policies in place.

   

Increasingly, data center data traffic is within the data center rather than from without. This increased intra-data center traffic is due to “rising use of IT services, remote storage, and the increasing use of real-time processing (enabled by high-speed user connectivity) such as mapping, voice recognition, industrial and medical diagnostics, and big data analytics.” These increases have resulted in peak demand characteristics similar to those of the electric grid, which necessitates the use of expensive “reserve capacity” for data, which translates to more overall energy use as more equipment is turned on and readied for use in the event it is needed.

Bandwidth is another issue facing consumers in crowded areas where bottlenecks are slowing connection speeds. Some places have over a million connected devices per square kilometer. More fibre optic cable is being built, including several new transoceanic lines. Data center are also linked with fiber-optic lines. New fibres that reduce noise and upgrade the hard limits of the old ones are being developed to increase data transmission speed. Processes like voice recognition and video require the higher speeds to run optimally. Autonomous cars and 3D virtual reality also require the high speeds. Networks are planned to go from old 2G, 3G, and 4G to much faster (up to 100 times) 5G in the 2020’s. Improved signal processing technology is enabling more independent data streams per cable. Optical processing may be a technology of the future that can solve speed issues but still has some hurdles to overcome.

   

Wireless networks are also big energy consumers (overall higher than data centers but not as high as end-users and manufacturing) with most energy going to radio power amplification. Cooling, data processing, and power supply also contribute. Here as in data centers, efficiency improvements have occurred but lag far behind increased data and energy usage.

At the end of 2012 there were 3.2 billion cellular subscribers with 1.2 billion of them being broadband also. Theoretically, there will be a slowing of growth in this sector as all 7 billion people on the planet get connected in this way! Of course, with industry and business subscriptions there will be even more. There are new more mobile devices than human population.

Apparently, many cell towers are powered with diesel generators but many are moving to the grid as they can. There may be an opportunity here for renewable energy-based microgrids as well. High speed LTE cellular networks increase energy usage. Mobile wireless data uses more energy than wired data and its use is increasing all over the world due to the convenience. All those people staring at their phones (me included) are consuming energy, are likely burning coal. In India some data centers are powered by diesel generators which emit abundant black carbon, or soot, a hazardous particulate matter.

Digital TVs and gaming, particularly on-line gaming, are also huge data users. Digital industrial scale 3D printing is also slated to increase and become a major data/energy consumer. Other end-use devices are huge data users as well and as more appliances and control systems become connected that use is set to skyrocket further. Embedded software and anything “smart” is effectively a significant consumer of data which is then a significant consumer of energy. While such devices often save home or industry energy use as they are advertised to do, there is often very little accounting of the increased energy use from increased data use. Such devices undoubtedly benefit the consumer and industry but there is a down side too.

  

It also takes massive amounts of energy to build digital ICT devices and those devices often end up with shorter lifetimes than other products. Building semiconductors in microprocessors use the most energy in manufacturing ICT devices. The world’s semiconductor industry, worth hundreds of billions of dollars, builds products that cannot be recycled. At the end of their useful life, semiconductors become trash. PCs, then smart phones and networks, make up the bulk of energy usage in ICT manufacturing. Industries, cars, and appliances are also being outfitted with semiconductors. 80% coal-powered China has about 50% of the global market share for semiconductor manufacturing. The Semiconductor Industry Association (SIA) put out a report at the end of 2015 that projected that as chips get more dense (more transistors per chip and more interconnects) then energy use will continue to go up since more interconnects means less overall efficiency. Perhaps the bottom line is that more speed means more power drained. Thus, the Moore’s Law trend in semiconductors of the past will begin to meet up with an energy use ceiling. New materials and new technologies would be needed to offset this problem. Another SIA report notes that by 2020 it will no longer be viable to decrease chip size by the conventional methods used today.  

  

Somewhat counter-intuitively, the increase of data efficiency and energy efficiency in data intensive processes does not generally lead to decreased energy use but often to increased energy use, as more processes become available to use for less expense as a result. This is possibly a case of the so-called Jevon’s effect, or Jevon’s paradox, whereby increased efficiency leads to increased overall consumption and not less overall consumption as hoped. It is a simple case of buying more of something that is cheaper. It is the increases in efficiency of microprocessors that has enabled the increase in global data traffic through making such hi-tech affordable. Thus, the well-known economic principle that ‘efficiency simulates demand’ is apt in this case. Efficiencies are still worth pursuing and there is an offsetting effect and some limits to them simulating demand, particularly in direct energy efficiency when it becomes a goal in itself regardless of cost, such as when including the social costs of carbon and pollution. There is still “low-hanging fruit” as many inefficient first generation servers built over a decade ago in are still running, mainly in older smaller data centers. As they are replaced energy efficiency will increase Given that the efficiency growth in computing according to Moore’s Law began to slow in 2005 and continues along that trajectory, it is fairly certain that data/energy use growth will outpace it. While there are other sectors where energy efficiency is indeed reducing energy use, the ICT realm is not one of them and is not likely to be any time soon.

   

References:

Smaller, Faster, Lighter, Denser, Cheaper: How Innovation Keeps Proving the Catastrophists Wrong – by Robert Bryce (Public Affairs, 2014)

The Cloud Begins with Coal: Big Data, Big Networks, Big Infrastructure, and Big Power – by Mark P. Mills, CEO Digital Power Group, sponsored by National Mining Association and American Coalition for Clean Coal Electricity, August 2013

My New Study of Data Center Electricity Use in 2010 – by Jonathan Koomey, July 31, 2011

Google Uses Artificial Intelligence to Boost Data Center Efficiency – by Robert Walton, in Utility Dive, July 21, 2016.

How Clean is Your Cloud? – Greenpeace International, April 2012

Making the Cloud Green: Tech Firms Push for Renewable Energy Sources – by Stephanie Joyce, from – All Things Considered, National Public Radio (NPR), July 22, 2016

Computers Will Require More Energy than the World Generates by 2040: Moore’s Law is About to Hit a Wall – by Peter Dockrill, in Science Alert, June 26, 2016

Here’s How Much Energy All U.S. Data Centers Consume – by Yevgeniy Sverdlik, at datacenterknowldege.com, June 27, 2016

The Problem of Inefficient Cooling in Smaller Data Centers – by Yevgeniy Sverslik, at datacenterknowledge.com, Dec. 4, 2015

New Mexico Reviews Power Plans for Facebook Data Center, in PennEnergy, Aug, 2016

The Bandwidth Bottleneck That is Throttling the Internet – by Jeff Hecht, in Nature magazine, August 10, 2016