Text Size AAA
Jun 11, 2013

Logged On, Tuned In, and Energy Efficient: Just how green is the Internet?


On Thursday, June 6, 2013, Google hosted “How green is the Internet?” The one-day, invitation-only summit brought together 100 experts from industry, academia, government, and NGOs to explore emerging questions around the environmental impacts and benefits of the Internet. The goal of the event was twofold: 1) expand the energy discussion of the Internet beyond just data centers, and 2) brainstorm transformative ideas for using Internet communications technology (ICT) to overhaul current business models in a way that creates transformative environmental and societal benefits.

This was the third event of its kind hosted by Google. In 2009, Google convened experts to discuss data center efficiency, similar to a 2003 RMI workshop that led to the publication of High-Performance Data Centers. In 2011, Google broadened the topic scope from data centers to the data center industry. The 2013 event continued that evolution and considered the ICT sector at a larger scale, looking at end-to-end energy and societal impacts of all aspects of the Internet, including:

  • end-user devices (only the energy associated with connecting web applications)
  • data centers where users’ data is stored and/or processed
  • core and metro networks that connect a user’s access network to data centers
  • broadband access technologies, including fixed broadband, mobile (base stations providing access to 3G and 4G networks), and local wireless (e.g. a wifi hotspot, a router in your home)

Through my haze of star-struck ardor, I listened (from the 2nd row!) to keynote speaker and former Vice President of the United States Al Gore frame the importance of the problem: the ICT sector and its associated energy use are growing at an unprecedented rate. Within the next seven years, there will be 50 billion smart devices connected to the Internet. Researchers estimate this could account for around 10 percent of current U.S. electricity consumption. Really, no one knows; not even Google. But the number will be significant. With that staggering energy consumption in mind, Gore stressed the urgency of the climate crisis we are facing, referring to the recent news reports of climate-related weather disasters as “a nature hike through the Book of Revelation.”

But while this event acknowledged and focused on the energy use and environmental impact of the ICT sector, it gave equal credence to the societal and environmental benefits provided by the Internet. Morning presentations tackled a number of key topics, including Internet infrastructure; energy use implications of cloud-based email, docs, and software; comparing traditional retail and e-commerce, and CDs vs. digital music; and how much energy does it take to send a gigabyte of data?; among other relevant subjects.

Eric Schmidt, executive director of Google, kicked off the afternoon breakouts with an inspirational talk that really got to the heart of the matter—the growth in this sector is going to continue. And it’s adding immeasurable societal value in ways that we never could have dreamed of. The Internet has provided consumers with information to make more sustainable consumer purchases and enabled telecommuting and teleconferencing as alternatives to carbon-intensive travel. In some societies, the Internet provides people with their only access to medical treatment, politics, education, and socialization outside of their culture. “You think the Internet matters?” he asked. “It matters a lot.” He stressed that the solution shouldn’t be to use the Internet less, and certainly not to limit the global reach of these services. Instead, we must make the system components more efficient; power the sector with clean renewable energy; and leverage the Internet to help solve the climate crisis.

Eric SchmidtIn addition to learning about topics that directly connect to RMI’s work, I was also excited to offload all of my million-dollar app ideas into the capable hands of people that can actually make them happen. Examples include a parking app that identifies shaded parking spots to keep cars cool, and an Industrial Ecology mapping program that charts waste streams across the U.S. to allow industries to identify suppliers or purchasers of various byproducts and waste.

Later, we divided into facilitated breakout sessions focused on e-commerce (banking retail), digitalcontent (newspapers, music, photos, video), communication and collaboration (email, social media, video conferencing), and Internet solutions (logistics, mapping) to brainstorm ways the Internet could be used to create transformative environmental benefits. Take the Streetline “Parker App,” for example. Consuming only 15 watts, this app allows you to find parking in an urban area, drastically reducing emissions from unneeded driving and idle time combing the streets for an available spot. The sessions generated a number of ideas, including:

  • using the Internet to scale crowdfunding for solar and other renewables
  • using mapping programs to optimize shipping and distribution across companies to minimize trucks traveling empty
  • an environmental “guilt meter” tied to social media that tracks consumers’ purchases
  • pairing a smart grid with geographically dispersed renewable resources to provide clean power where it is needed—a concept near and dear to RMI!

Top 5 Things I Learned

Event PhotoOverall, this was one of the best professional events I have attended throughout my career. Hats off to Google for putting on a truly special event and bringing together a diverse and passionate group of individuals. (Those individuals even included three other RMI connections: RMI alum and director of sustainability at Yahoo! Chris Page, and RMI senior fellow and research fellow at Stanford University Jon Koomey, pictured with RMI alum and Google data center sustainability analyst John Stanley.) As I continue to think about what role RMI should be playing in this space, I reflect on the top five things I learned at this event (in order of importance/interest):

1. Energy use of wireless access networks dwarfs that of the device and the data center for mobile devices. By 2015, the wireless cloud will generate 30 megatons of CO2—90 percent of that comes from the mobile access networks!

2. Lawrence Berkeley National Laboratory has recently developed CLEER, the Cloud Energy Emission Research model, to calculate the environmental benefits of moving applications to the cloud. This model is, in my opinion, a key tool for making a compelling business case to facility and IT managers to migrate to the cloud.

3. Cloud-based software could reduce current data center energy by as much as 85 percent. The technical potential of this savings is 23 billion kWh/yr, roughly equivalent to the annual electricity use for Los Angeles, CA. Refer to my recent presentation with Mark Monroe (DBL Associates) and Josh Whitney (WSP Group) on reducing data center energy use for more information.

4. For ultra-low-power devices such as smart phones, tablets, and low-power laptops, production (not operation) accounts for the majority of the lifetime emissions.

5. One of the most powerful benefits of the ICT sector is dematerialization—moving bits and bytes of data instead of pounds of people and products. The net environmental impact is drastically lower for downloading music or an e-book versus driving to a store and purchasing a physical good. In case you’re curious, downloading an e-book or audiobook, which accounts for more than 90 percent of its energy, consumes about 0.6 Wh of energy, the equivalent of powering a 100W light bulb for 20 seconds.

How would you like to see RMI get more involved in reducing this growing energy use? How do you think the Internet can be used to solve environmental problems?  

Recommended Reading 

Image 1 courtesy of Shutterstock.com. Al Gore image courtesy stocklight / Shutterstock.com. Eric Schmidt photo courtesy of Google.com. Event photo courtesy of Kendra Tupper.

Join the Discussion

Showing 1-10 of 11 comments

June 12, 2013

Sounds like a fascinating summit. Great post Kendra!

June 13, 2013

Just checking this out and not having read your references, there was a recent two part NY Times article on the very same issue, but they contend that the industry is staying buttoned up about most of what happens.


An estimate I read was that most servers are only running at 30% capacity. There are also some interesting independent studies about the subject of energy use and the cloud:

One article I cannot find right now is about a European study monitoring energy consumption on desktops/laptops, tablets, and books for reading long form copy. I believe they found that even books are GHG competitive, though the original Amazon Kindle reigns supreme.

Lastly, I also contend (as I believe Google first put forth) that good site coding and properly compressed images play an ever increasing role too. What are your thoughts? I'm really interested in this.

-Noble Cumming (freshly minted Sustainable Graphic Designer with an MA in Sustainable Design from MCAD's Sustainable Design Online program).

June 13, 2013

Thanks for great article and the 2 links "how much energy does it take to send a gigabyte". I did some math before on Koomey's calculations (used in the buy.com paper) of 9-16 kWh/GB (which I find hard to believe) and the reference above at 1.6 kWh/GB. While data center cooling efficiency is a fashionable topic, I don't think people realize how much energy is lost in transmission. Do the math-- data centers need an order of magnitude less electricity than transmission. Here are some numbers...

Let's use the Coroama estimate of 1.6kWh/GB-- a 50W server at moderate efficiency could deliver 100Mb/s, but the "internet" energy required comes out to be 72KW! An inefficient cluster at 1kW could easily deliver 1Gbps, but consumes ~1 megawatt in transmission. A 1.5Mbps DSL line uses 1kW if fully loaded.

The paper on buy.com has some math errors, but they assume 1MB/order of internet traffic. Well maybe in 2008, but just the buy.com home page is 1MB. If I access a 1MB page every 30s (I comparison shop), that comes out to consuming 192W while shopping-- many times more than my laptop & router/modem.

But the moral is RMI (and us) should be promoting "web efficiency" not just data center cooling efficiency. Take this page (before my comment)-- it's 2MB, and way too large. I used firebug to look at the content and see some silly mistakes. For example, the Most Recent Blog Posts thumbnail of Franconi is full size at 694KB instead of a 1650 pixels. There are also some massive javascript downloads. The HTML of 74KB would be 27KB if sent compressed, but is still bloated as the text it creates is only 16KB. So with some better software, you could cut your data center and internet energy by 90%-- and that doesn't even get into more efficient computers or air conditioners, etc.

The buy.com site is pretty bad as well-- not compressed and .6-2MB per page, with only 10s of KB of text. Amazon is more efficient, but still has quite large pages.

I'm not sure how many readers access this page, so maybe you wouldn't care about eliminating a couple of kWh. But why should shopping or scanning blogs take 200W instead of 20W or even 2W? Why do I have to wait 5-20s for pages to load? If I bought a 50Mbps high-speed internet, then would that just give web designers license to make their sites use 20MB/page and up the hidden internet power to many kW/user?

It would be interesting to have an EnergyGuide icon on a web site and show annual internet electricity cost while reading news online, along with a bloatware meter showing what it could be. I believe data center and internet energy are 10x larger than needed just by eliminating waste and creating pages in a more sensible way.

For web server efficiency, it would be interesting to have a contest that measured web pages served per kWh, say with some open-source realistic data like Wikipedia. It's a hardware, software, and data coding problem, and I would winners would be several orders of magnitude better than what is typical now.

June 14, 2013

Carl, You are off by close to three orders of magnitude, and thank goodness for that!

I recall reading a decade ago of "breakthrough" ADSL line drivers that consumed only one watt per line. Newer and faster DSL line drivers do consume somewhat more power, but not orders of magnitude more. Transmission power is typically a fraction of a watt, with heat loss in the modems at either end the only significant power sink. Now DSL range is only the "last mile" to a premises, but the power efficiency of long-haul links is of necessity much higher. A trans-pacific fibre optic transmission facility with close to one terabit per second of data capacity does not draw a whole megawatt of power.

(See the table on pages 11-12 of this pdf, published as supporting material for the summit : http://static.googleusercontent.com/external_content/untrusted_dlcp/www.google.com/en//events/howgreenistheinternet2013/pdfs/DirectEnergyDemand.pdf )

June 14, 2013

Spam uses 1% to 2% of all electricity!

depending on how many servers it propagates.

modern filters may have reduced this below 1% now, but not sure.

At one point 80% of all emails where spam

June 14, 2013

SIC power supply designs need no cooling.
If relatively hot waste heat cooling exhaust from servers is routed past the power supplies the heat will not contaminate the silicoln server farm components with radiated heat.
Cool climate building heat can be obtained from this waste heat recovery also in northern climates for much of the year.

The bandgap is 2 to 2.2x of Si so not going to be advantageous to build computers along with the high degree of crystal defect compared to silicon.
Prices are coming down and when payback period is considered for cooling cost reduction potential and multipliers for designs listed above are already cost competitive. Under 3 years with standard design, under 1 with novel series, high density cooling design like the one listed above.

June 14, 2013

If ad companies where forced to pay for their end to end energy use the ads would be cut to 1 to 5% of the total now because it simply would not be cost effective.
Sam for spam.
Our internet access bill pays fro this crapola that is unwanted.
With corporations footing ~ 2/3 the advertising bills for this stuff when their employees surf for work and personal internet time.

I can block adds with Firefox add-ons but that only stops them frombeing displayed on my screen, not from propagating through the dozen servers and redundant networks that provide me with this unwanted, unasked for crapola.

June 14, 2013

A couple of critical thoughts for folks:

1) According to the SMART2020 report, ICT (Information and Communications Technology) *SAVES* more than five times its carbon footprint in the rest of the economy. Increasing to 10% should be a good thing as that implies substantial and disproportionately more carbon savings by doing things smarter and more efficiently than without.

2) The NY Times Article, below was roundly criticized, as not only sensational and inaccurate, but completely lacking in context. The actual energy use by Data Centers is 1.3% worldwide and 1.7% in the US. Moreover, all the cell phones in the world use more energy just to recharge their batteries every night than all the servers in the world.

3) When you look at the Pareto Chart of where our energy is used, you have to come a *LONG* way down to get to "The Internet". Transportation, Heating & Cooling, Dryers, Stoves, Lighting, Bulk Generation and Transmission are double-digit uses that we shouldn't forget about optimizing in our haste to go after the next shiny thing dangled in front of us by a narrowly drawn look.

Let's keep our focus on the big things, first.

June 14, 2013

Johnathan, thanks for the prompt to look at the conference link. Your URL was chopped but I found and played with the LBL CLEER model-- really a neat tool comparing local servers vs cloud based computing. A main point of this site is to show benefits of cloud computing with shared virtual servers vs individual machines mostly running idle-- I believe in that efficiency. (But why does a server take 250W while my equally powerful laptop with display uses 13W while reading, peaking at 25W loading javascript? Plus a 3W pogoplug can serve 100Mbps-- as long as it's using old-fashioned html instead of PHP.)

So what is the kWh/GB? The CLEER site crashed when I tried it (500 server error) but with default data in CA I looked at Network energy/yr and GB/y, and converted to kWh/GB and came up with 1.03 kWh/GB Network, or 1.28 kWh/GB with Embodied Network added. That's about the same as Coroama estimate of 1.6-- which was for a longer distance.

I am still interested in a better reference for internet energy use-- it still seems high and also dependent on distance. In Koomey's paper he takes agregate telecom industry energy consumption. In Coroama, you can see a lot is in the chain of routers and connections near the endpoints.

Suppose RMI's blog server moves to cloud computing as proposed, and gets combined with other blogs and combined spits out a 100Mbps load for 1 hour, or 45GB. The CLEER model predicts the network energy to be 58kW for that cloud server output. (An efficient server could output 1000Mbps, so 100Mbps is realistic.)

I don't understand the CLEER default (California) cloud data center usage of 2994TJ/y or 2.5kW/GB transmitted. That's just what the defaults auto-set and comes out to only .86GB/h or 1.9Mbps traffic/server. So in the overall graph Network energy use isn't so big, but that's because the servers are (still) barely used.

What I was trying to say was that better software and design guidelines could make web sites 10 times leaner (and faster), so can cut both data center and network energy consumption. Sure spam sucks and transportation consumes way more, but no reason to ignore 10x bloat in online consumption.

A blog with page sizes such as this with 10000 users that read 10 pages per day will use $3850/y of network electricity with Coroama's figures, but might only need $500/y if the bloat was fixed. OK, most blogs aren't that popular, but think about it in aggregate. About 10 such sites would fit in my assumed cloud server @100Mbps, or 100 sites/server if trimmed down.

June 14, 2013

Carl -
If your server is using 250W at idle, trade up to something built this decade. The servers that the 2007 EPA report talk about took 225W at idle, but a modern server only takes about 55W at idle, today. And in the mean time, the performance is significantly better, and the energy use is proportional to how fast it is running. Thus, it no longer maters if a server is only 30% loaded, because it's only using 30% the power of full load. Unlike the old servers that used 225W at idle and 400W at full load, modern servers have an almost linear energy use with performance these days. In fact that 225W pretty much runs today's server at 100% and produces 50 times the work the old one did at 400W.

PAGE: 1 2