Skip to content

An Energy Efficiency Parable

Here’s a story that captures a lot of the challenges we face as we try to improve energy efficiency. Spoiler alert: it doesn’t have a happy ending, but I’m holding out hope for the sequel.

Many of us leave our work computers on 24/7 – estimates suggest more than half of us. I do mainly because I like to use remote desktop from home.Computer on at Night

In some organizations, the computer teams insist that desktops remain on overnight so they can run software updates, download the newest security patches, backup files, and do other important stuff during hours when people are less likely to be working.

The folks who pay an organization’s energy bills – let’s call them the accountants – would probably be unhappy if they knew how much they were paying to power computers between the hours of 5PM and 9AM. It’s a lot of money — at electricity prices of 11 cents per kilowatt-hour, a 75W computer left on for all 8760 hours costs almost $75 per year, while one left on for 2000 working hours only costs $16.50. Estimates suggest that nationwide, desktops consume several times more energy than servers.

But, accountants have no visibility on the computers’ energy costs – they simply pay a utility bill at the end of every month and few have any idea how much of it is going to computers versus air conditioning, lighting, etc.

Even if accountants knew exactly how much money they were spending on idle computers, it’s unlikely they could do anything about it. The computer folks, who operate the network, and in many cases dictate what kind of computer employees can buy, would revolt if the accountants mandated that computers be put to sleep or turned off at night.

Sleeping ComputerThe computer team cares much, much more about keeping the computers virus free than about saving energy. They will get midnight calls or hauled in to see the boss if anyone is infected by a virus.

Employees probably would be upset if they had to turn off their computers, too. How annoying is it to run in late for a phone call and then have to wait 3 minutes for your computer to boot so you can find the phone number? This isn’t an issue if computers are put to sleep and not turned off. But, many of us don’t know how to do that, have bad associations with early versions of sleep, or just can’t be bothered to do it.

Like other energy efficiency examples, this is a case of split incentives – the computer teams’ and employees’ versus the accountants’ – plus poor information.

No computer team is going to request computers that use less energy if it in any way makes the computers more vulnerable to viruses or harder to operate. And, no one in the organization has good information about how much money is spent on energy because of nocturnal computers.

But wait, my neoclassical economist friends might say. Isn’t there a boss who cares about BOTH lower energy bills and productive employees with virus-free computers? Shouldn’t the boss be motivated to collect the information and then get the accountants and the computer team to sit down and figure out a good solution? Workdays would need to be 108 hours long before that crossed most bosses’ minds.

What if Apple and Microsoft develop operating systems that allowed computers to get downloads off the network even while it’s asleep? Computers that are asleep use almost no energy – 1-2 Watts – it’s almost as good as turning them off.

A colleague of mine at LBNL, Bruce Nordman, was part of a team that set out to solve just this problem. If this were a movie, they’d be wearing capes.Superhero at Computer

They worked with a standards organization and formed a committee that included Apple, Microsoft, AMD, Intel, Sony (game consoles are networked and often stay on to receive occasional bleeps), among others. This is no easy task to get a bunch of competitors together to talk about collectively changing their products, particularly when the topic is as unsexy as energy efficiency.

They helped develop a “network proxy,” which effectively lets a computer go to sleep while the proxy tells the network it’s awake and then wakes the computer up if anything important comes down the pike. Energy Star now provides an incentive for companies to include the feature in their product, and the program’s early support for the technology was critical to initiating the standards process.

Apple bought into it, as did some chip manufacturers. Unfortunately, Microsoft didn’t buy in. And, Apple now ships its computers with the capability turned off, though you can turn yours on – see here, it’s called “Wake on Demand.”

What are the general lessons about energy efficiency? First, we should encourage more teams like Nordman’s. How? We can’t rely on the market to support them because split incentives and information asymmetries are examples of market failures.

Public research and development funding is certainly one answer. Nordman and his team were funded by state and federal grants. And, while the network proxying story has yet to make much progress, they have had other successes, like encouraging Ethernet chip providers to change the standard to use 75-90% less energy.

A lot of people are championing energy efficiency. The recent Shaheen-Portman bill on energy efficiency had bipartisan support, though it eventually failed. Most carbon mitigation scenarios assume a hefty share of the greenhouse gas reductions will come from energy efficiency. But, figuring out how to unlock energy efficiencies’ potential will be difficult. And, it’s less exciting than developing a new solar panel or a new gizmo.

Fortunately, there are a lot of smart people, like Nordman, trying to tackle the organizational, as well as the technical, issues in order to solve the energy efficiency puzzle.



Catherine Wolfram View All

Catherine Wolfram is Associate Dean for Academic Affairs and the Cora Jane Flood Professor of Business Administration at the Haas School of Business, University of California, Berkeley. ​She is the Program Director of the National Bureau of Economic Research's Environment and Energy Economics Program, Faculty Director of The E2e Project, a research organization focused on energy efficiency and a research affiliate at the Energy Institute at Haas. She is also an affiliated faculty member of in the Agriculture and Resource Economics department and the Energy and Resources Group at Berkeley.

Wolfram has published extensively on the economics of energy markets. Her work has analyzed rural electrification programs in the developing world, energy efficiency programs in the US, the effects of environmental regulation on energy markets and the impact of privatization and restructuring in the US and UK. She is currently implementing several randomized controlled trials to evaluate energy programs in the U.S., Ghana, and Kenya.

She received a PhD in Economics from MIT in 1996 and an AB from Harvard in 1989. Before joining the faculty at UC Berkeley, she was an Assistant Professor of Economics at Harvard.

12 thoughts on “An Energy Efficiency Parable Leave a comment

  1. What a brilliant article on the “pleasures” of my daily job since 2007. I agree that the main problem is the “I don’t care about saving in a field out of my budget that also could disturb my job” (the IT Staff’s vision) jointly with the impossibility for any other area to find out if to Power Manage the computers will really make SAP, Oracle, the storage and the email server to explode overnight (the truth is… they will not).

    I’m seeing some tendencies changing, and gladly IT people is more engaged in this adventures because maybe time has arrived. Also, as 100% of energy managing and metering initiatives in companies have to deal with IT departments, the friction makes them friends with FM and other general services in buildings.

    Regarding the WOL and proxies. I work with Verdiem’s software, and we use a “role based proxy” with zero network configuration. What does it mean in english? We install an agent in each PC (it is really fast, like 2000 can be deployed overnight). This agent has 3 roles: The first one is to meter user’s activity and PC power states (so waste means no activity and energy intensive state), the second one is to store and apply energy saving policies (they are cleverly applied when the user is not affected because we definitely know when he is not there) and third, it can adopt a proxy role in an automatic way: we tell the system to select a given number of proxies for each subnet (usually 2) and some of the agents automatically make the given number of PCs awake 24/7. This has negligible energy impact in big networks and this computers are alert “hearing” for incoming orders to wake one or sets of peers in that same subnet (the orders arrive via HTTP and port 80 because the system is a web based one, so there is ZERO ports and things to open on firewalls. This is the most efficient way of being able to maintain computers suspended or off while you can also turn them on (or activate them from standby) at wish on demand or in a scheduled way. When I say most efficient I mean that it is set up in 10 minutes, works 100% of the time in standard networks, you don’t have to ask for absolutely anything from the network managers, and you don’t have to manually configure anything.

    I like to explain this functionality because I’m glad to be the seller of the biggest PC Power Management project up to date in Europe, affecting 35.000 computers in a Gov Agency in Spain, where the biggest reason to buy from the customer is NOT to save energy, but to improve network management because of this WOL proxies.

    I’d like to share this history explaining my vision on this area:

    Here you are a Wake On LAN readiness guide document from Verdiem

    And some thoughts on why some basic tasks are enough without going nuts with policies in PC Power Management.

  2. Ah, but my university has this thing called the New Berkeley Way of self improvement and we have at least a couple thousand computers (legacy towers) that run 24/7/365 that could be addresses by a small team of equivalent FTEs of IT folks.

    So does every large corporation or firms

  3. The notion that with server virtualization the Chief Information Officer, rather than the Chief Financial Officer, is the residual claimant on electricity bills and thus incentives are better aligned (the point made by art2science), is perhaps worth exploring. As I read Catherine’s article, one thought I had was precisely this. In owner-operated firms, with the bossy owner playing both CIO and CFO and much more, my guess is that most office desktops are put to sleep overnight. This suggests that the cost-benefit analysis proposed by Paul Brooks may not indicate that the optimal solution is to keep desktops powered up overnight.

    Home desktops offer another angle from which to view this outcome. How many of us high-value-of-time workers really keep our home desktops on during the day while we are at the office? The value of 3 minutes while at the office is probably similar to 3-minutes worth at home, at least for white-collar workers.

  4. My Windows 7 desktop PC has a “Wake on LAN” setting precisely to let administrators do stuff on maintenance etc. Since I never access it remotely, I’ve never activated it, but there it is.
    Someone should respond about Windows 8, which I haven’t been interested in (old programs working in Virtual XP mode). So I don’t understand why it was necessary to create a network proxy, unless it is for computers that are off the main network. Indeed, I had to modify my sleep mode at home to turn off “wake on mouse” because the heavy trucks and buses going by often rattled my front room study desk enough to tremor the mouse.

    • Gene,

      Wake On LAN and network proxy are subtly different concepts. Under WOL, a machine goes to sleep and all activity ceases. The Ethernet interface (it is an Ethernet hardware protocol) watches for a “magic packet” with specific encoding including the machine’s hardware address, and if it is detected, the machine is woken up (or turned on). Once the computer comes to its senses, a network administrator can connect to it and “do stuff.”

      Under the network proxy concept, the machine sleeps, but in essence, a smaller, simpler low-power machine continues to operate and accept network traffic on behalf of the sleeping machine. With this, various bits of network state can be maintained and some network activity can be attended to without involving the main machine. To the outside world, the machine appears to be awake all the time, even though in reality it has been sleeping, or has been woken only periodically.

  5. I noticed that “wake on demand” setting on my Macs, but I 1) never knew how it worked and 2) did not realize it was an attempt to be an industry standard. This problem used to drive me crazy – I once found an entire empty computer lab of desktops + CRT monitors that was left running for the summer, because of the IT motivation you cite. Given your explanation, I will start to use wake on demand.
    I thought notebook computers were going to solve this problem – my colleagues don’t need to leave our office computers running, because we take them home with us. But that’s a minority of enterprise employees. Another potential solution is in the microprocessors and GPUs. They have multiple levels of “sleep states,” and can be set up to drop to a lower power level after non-use in a variety of clever ways. But as you suggest, that requires both the OS (to create hooks for it) and the IT staff (to use those hooks). It could be implemented at the motherboard level, as well. Dell and a few other vendors should use this as a selling point.
    One more factor adds to the cost – air conditioning loads can be larger than the direct electricity consumption. That $75/year electricity bill can be $200 in some situations – which in a few years approaches the cost of the computer. In offices without AC, leaving your machine running all night/weekend makes your office appreciably warmer the next day.
    David: my understanding is that server virtualization has greatly increased the load factor, and therefore power efficiency, of servers. Because servers are paid for out of the IT budget, incentives are aligned.

    • Virtualization is certainly one way, though not the only way to increase server utilization and therefore reduce the number of servers required to handle some particular load. If reducing costs were the only goal of IT organizations, then perhaps, yes, there is broad alignment between capital costs (borne by IT) and the operational energy costs (borne by facilities or whoever).

      However, virtualization has downsides. For example, virtualization and reliability are not friends. You obviously gain nothing in reliability, at least with respect to hardware problems, by duplicating virtual servers on the same physical machine. Moreover, if you put too many different virtual entities on one hardware device, then a fault in that device could ruin your whole day.

      My point being is that if price were no object, most IT shops would probably prefer not to virtualize. At best it complicates their planning and creates subtle performance interactions between services that might have been completely independent before being combined onto the same hardware.

      Price /is/ an object, so incentives are somewhere in between, I suppose.

      Regarding CPUs and sleep states, I think most OSes take pretty good advantage of the low power states of their CPUs. However, the CPU is a surprisingly limited fraction of a server’s energy use. (GPUs are not common in servers though they appear in CUDA cluster and rendering farms, etc, both pretty niche-y today.) A server’s memory must remain powered, as well as its network interfaces. Connected hard drives may be powered down when idle, but be careful, powering them up can take seconds — eternity in server time. 🙂

      Excellent point about energy cost and cooling. The industry refers to the energy overhead to run IT equipment in terms of Power Usage Effectiveness (PUE). A facility with a PUE of 2 means that for every Wh of IT equipment there is another Wh of cooling, heating, power conversion, etc. PUE of 1.0 means that there is no overhead at all. State-of-the-art industrial data centers achieve PUE numbers on the order of 1.02-1.15, but your typical small office server closet is closer to 2.5-3.5. So, a watt of avoided IT load really saves 2.5x to 3.5x as much.

  6. While I completely agree that that increased energy saving is a good idea, it should be done with cost benefit analysis. $75 per employee per year is not a lot of money, it is negligible. For most business $50, $100 or more per hour is a typical cost, so savings $75 per year is just not significant. Assuming an employee works 250 days a year it is $0.30 per day. Certainly if the cost is low and not too much time has been spend on this, so there is a proven cost benefit analysis, then this is a worthy goal.

    One of the main problem with energy use is the production of greenhouse gases. As MacKay points out in “Sustainable Energy Without the Hot Air” available free on line, every little bit of energy savings adds up to a little bit overall. MacKay points out that every big helps. It is only big changes in the way that we produce and use energy that will make a difference if we are to reduce human greenhouse gas production from 20 to a sustainable 1 tonne per year.

    So saving the energy that computers use may be worthwhile if there is a favorable cost benefit analysis. However it is not significant enough to make any difference in overall energy use in N. America.

    • I think the original point was that even if a good CBR can be assumed, you still may not get the behavior. Assume for a moment that enabling sleep on every machine in an enterprise has zero-risk or inconvenience associated with it. Cost: zero, benefit: nonzero. In a split incentive environment, that hardly guarantees adoption.

      However, in reality enabling sleep is not costless, so the question is indeed a bit more complicated.

      The way I think of this sort of thing is to consider how X-intensive the business is. If your business is energy intensive, you look at saving energy because that’s a big bill. If it is water intensive, you naturally look at water. Will a law firm with 30 desktop PCs and four servers in the closet look at energy? They’d be crazy to. Sadly, most of the IT energy use is not in big data centers, it is in the small business with the closets, so we do have something of a problem!

  7. Why do accountants care less than IT staff about keeping the computers virus-free? We can just say that although accountants would be willing to pay more for computers that use less energy, that preference is not getting communicated back to suppliers at the purchase decision. At my university you see the split incentive there: employees get to pick their own machines without having differential energy use charged to their research accounts.

    • Agreed. Accountants get hauled in too when one exceeds utility cost projections.

  8. Great article, and I’m similarly dismayed that network proxy hasn’t been adopted. Despite the apparent risk-aversion of IT managers and Microsoft executives, I’m optimistic that we will see big gains in computing efficiency in the coming years; there is secular change in computing that may obviate the sleep issue entirely.

    For one, laptops are replacing desktops. Not only do laptops draw less power in all modes, they also obviously spend more time asleep. Worldwide CAGRs on desktop shipments, I believe, are negative these days, so that whole part of the industry is shrinking.

    Moreover, many believe that laptops will themselves be supplanted by mobile devices like phones, tablets, and other “lightweight” devices. This is fantastic news because all of those devices have fundamentally different “DNA” than the PC platform. They evolved in a highly energy-constrained environment. The result that all such devices sleep almost all the time, and you don’t even notice! (They also use low power processors, memory ,etc.) A phone is asleep unless you wake it to do something. It wakes essentially instantly and goes back to sleep as quickly. And, of course, it’s always ready to take a call or receive a text while asleep, the very essence of network proxy.
    Just as with desktop computers, energy efficiency for its own sake is probably not a serious concern of the device makers, but at least with these devices EE is free-riding on manufacturers’ goals rather than working across or against them.

    There is also middle class of devices, the “dumb terminals” of the 21st century that might be interesting to mention as well. Of these, I’m thinking of products like Google Chromebooks or thin clients that access a real or virtual PC over a network. These are low power devices, they can sleep or be powered off in seconds without consequence, and IT finds them easy to manage.
    Now, it is true that all of these types of devices interact with servers in “the cloud,” another source of energy use, but this is good news, too, because servers are much more efficient than office- based desktops because they get higher utilization (they simultaneously server many users) and because they are housed in facilities designed to operate them (primarily, cool them) very economically.

    There is a fly in this ointment: network power. As long as network energy use (the energy used to move data between devices and servers) is not very high, loading up a remote server with multiple users’ activity is a strong energy win. However, there is research ( that indicates that network power, particularly for wireless access (3G, 4G, etc), will eat up most of those gains. I haven’t had a chance to dig into this research, but it’s interesting and probably deserves attention.

    Two postscripts:

    1) There are some companies that have made progress cracking the split incentive problem for sleep mode, at least at the enterprise level, by offering products that the IT folks want (for remote management, etc) but which also manage sleep settings. The “accountants” still don’t get much say, but the IT people, getting the control they want, seem to feel a bit less risk-averse about sleep. Look at Verdiem, Verismic, Faronics, 1E, and others.

    2) Not exactly relevant to this piece, but it inevitably comes up in discussions of “wasteful” IT energy use: it’s probably unfair to compare the energy use of a fleet of desktop PCs to a fleet of wireless devices + data center + wireless network because the latter provides so much more utility — and is almost certainly enabling broad efficiency improvements throughout the economy, likely many times the size of its own footprint.

%d bloggers like this: