Taking care of the environment is an issue in which I've long had a strong personal interest. While I'm far from an activist, I try to follow practices that do the least harm to the environment. When it comes to library computing, good environmental practices also make solid economic sense. This month, I'll offer some thoughts on ways to reduce the amount of energy consumed by library computers.
When I first got involved with computing 20 years ago, one of the big questions was whether PCs should be turned off at the end of the day. One school of thought said that computers would last longer if left on. The argument was that the power spike associated with the initial start-up was more harmful to the internal components than the wear associated with longer operation. Others argued that turning equipment off when unused would extend the life of a computer.
Although this issue was subject to debate back then, in recent years the best advice suggests turning off computing equipment when it's not in active use. The components in current equipment suffer no ill effects from being powered off and on, and organizations can save energy costs by turning them off.
The High Cost of Power
World events tell us that the days of cheap energy are in the past. It was indicative of the time that the question of whether to turn off computers focused on the impact on the equipment more than on the cost of the energy involved. Times have certainly changed. Computer equipment costs much less and power costs lots more.
Reducing energy consumption not only makes sense from the environmental perspective, but it's increasingly a major budget concern. One could also say that rising energy costs demand power reduction, which, incidentally, has an environmental benefit. Either way, let's consider the methods of managing computers with less wasted power.
The energy issue hit close to home this year. While most libraries do not pay their utility bills directly, due to some of the oddities of how budgets are handled here at Vanderbilt, we do. The sticker shock of this winter's energy bills caused considerable worry about the budget. Some cost-saving measures were put in place, including postponing the creation of a new technical position to assist me with some of my projects. With the new budget year, things are back on track to fill this position, but it goes to show that energy costs can no longer be taken for granted, even in libraries. An economy in which energy costs represent a higher proportion of the library's budget provides strong motivation to reduce expenditures in this area.
For companies that deal with computing on a large scale, power costs are a surprisingly large component of their operating expenses. In the article "The Price of Performance" (ACM Queue, vol. 3, no. 7, September 2005, http://acm queue.com), Google engineer Luiz André Barroso provides some interesting insight into the economy of computing operations on the largest scale. He says that the four main factors in the total cost of ownership (TCO) for large-scale computing clusters are the costs for hardware, power, software, and the operation of a data center.
Google's strategy to create its own software and to take advantage of what's available in the open source community greatly reduces software expenditures since the company avoids paying the licensing fees associated with commercial operating systems, databases, and other applications. With software costs mitigated, hardware, operations, and power become the major expenses of Google's large-scale operation. The article says that while the company's hardware delivers increasing performance for roughly the same cost over successive generations, power efficiency remains flat. According to Barroso: "The result of these trends is that power-related costs are an increasing fraction of the TCO. ... If performance per watt is to remain constant over the next few years, power costs could easily overtake hardware costs, possibly by a large margin." In other words, it could soon cost more to power servers than it does to buy them.
Some Practical Advice
In order to save power, PCs need to be off when not in active use. That's no longer in question. This practice reduces energy consumption, saves money, and extends computer and monitor life. At a minimum, work computers can be powered off at the end of the day. But that's just the starting point. It's also possible to reduce power consumption during the course of the workday.
While it's common to employ screen savers that kick in when a computer is idle, this can be counterproductive to an effective power-management strategy. Screen savers do nothing to help reduce a computer's power consumption. Older generations of CRT monitors were subject to burn-in if the same image remained on the screen for an extended period. In other words, the phosphors on the inside of older generations of CRTs could retain the shadow of a static image displayed for very long periods of time. To avoid this problem, folks employed screen savers that produced constantly moving images when the computer was inactive. Their use became ubiquitous on PCs.
More recent CRTs are less susceptible to burn-in, and flat-panel LCDmonitors do not exhibit it at all. However, the use of screen savers continues, often under the misconception that they have some benefit. In addition, screen savers do nothing to reduce the computer's workload. Some have fairly intensive graphics and actually require more computer processing than spreadsheets or word processors. In some ways, screen savers can be harmful because they keep the computer user from enabling features that can actually extend the life of a monitor and reduce power consumption.
Screen savers solve a problem that no longer exists. Today's energy issues require a strategy that decreases the amount of energy consumed by computers. Most recently manufactured PCs include a set of advanced powermanagement features that can dramatically reduce the power these computers use. This power-management scheme automatically steps down the computer functions based on intervals of inactivity.
The Environmental Protection Agency and the Department of Energy sponsor ENERGY STAR (http://www.energy star.gov), a program that provides incentives and guidelines for reducing the power consumed in homes and businesses. To be eligible for the ENERGY STAR label under current guidelines, computers need to offer a standby mode that consumes less than 10 percent of the power supply's capacity. Almost all computers that are currently manufactured meet this standard, but the EPA estimates that only about 5 percent operate with the standby mode activated.
Anew set of ENERGY STAR guidelines, slated for adoption in 2007, goes beyond requiring a sleep mode. The new standards specify that the power supply operate with at least 80-percent efficiency, with some additional requirements on the levels of power consumption when in active operation. While it's good to purchase computers that are compliant with ENERGY STAR, it's also important to understand and take advantage of power-management features.
In the past, computers were either on or off. Now PCs with power-management features have multiple levels of activity. When in active use, the computer is fully operational-the monitor is on, the disks spin, and the processor is hard at work. This is when the computer draws its full level of power.
Computers can be programmed to enter standby (or sleep), a lower-power mode that follows a period of inactivity. In this mode, the computer shuts down its processor, disks, and monitor and then waits. As it enters standby, the computer just stops where it is without closing programs or saving any data currently in memory. With the touch of a key, the computer instantly awakes from standby and is ready to resume work. Standby works well for short intervals of inactivity. The danger is that an abrupt loss of power could result in lost data because the contents of memory have not been saved to disk. Standby uses some power, though, since a remnant of the computer's functions has to remain active in order to monitor the keyboard and mouse and to keep the contents of memory refreshed.
Hibernation is a deeper level of inactivity that draws almost no power at all. To hibernate, the PC leaves all the programs in their current state and keeps the user logged in, but it writes all the contents of memory to disk. The process of going into hibernation can take a minute or two, and it takes just as long to revive the PC. Hibernation makes sense for longer periods of inactivity.
Shutdown differs from hibernation in that it closes all applications, logs the user out of the workstation and the network, ends any background processes, gracefully exits the operating system, and powers down the computer. It usually makes sense to perform a full shutdown at the end of the workday.
To take advantage of this staged stepdown of activity, you need to configure your computer accordingly. While laptop PCs may come with these powermanagement features already enabled in order to conserve battery use, desktop PCs may not be set to use powermanagement features by default. It's easy to check and modify these settings. With Windows XP, for example, you can right-click on the desktop, select "Properties," and click on the "Screen Saver" tab. At the bottom of that panel will appear a button labeled "Power," which when clicked will display the "Power Options Properties" page. There, you can set the interval at which the PC should wait before turning off the monitor and hard disk. You can also decide when the PC should enter standby mode and when it should hibernate.
For most desktop computers that perform no server functions, I would recommend choosing values that turn off the monitor after 5 minutes, turn off hard disks after 20 minutes, enter standby after 30 minutes, and hibernate after 45 minutes. Such settings will achieve fairly aggressive power savings with minimal inconvenience. (Rememher that waking up from everything other than hibernation is practically instantaneous.)
Turning Up the Heat
Another component of computerrelated energy consumption is heat. Computers often generate heat more efficiently than they perform their assigned work. It's not an insignificant factor. Though you may not notice the difference in an office that has a single computer, the effect becomes more pronounced when you consider the total number of computers employed within a large building or campus of buildings. In rooms that house a lot of PCs, the generated heat has a much more immediate effect.
In one of the facilities I manage, the amount of computing and video gear had increased to the point that we had to install additional cooling equipment to supplement the building's central air conditioning. While dedicated cooling equipment is standard in data centers, it's fairly unusual in staff work areas. The compounded issue of using energy to power the computer and even more to cool the work area demands the selection of efficient computing equipment.
Server Management
Servers, unlike desktop and notebook PCs, need to be on all the time. You can't achieve good performance if the server has to wake up from a standby state before responding to a request.
This doesn't mean that reducing power consumption on the server side isn't an important issue. As I noted above, organizations that manage large numbers of servers give a great deal of thought to the power they need to operate their data centers. Power savings come more in the selection and deployment of server equipment. Rack-mounted servers, for example, tend to operate more efficiently and use less power than free-standing towers.
It's also not necessary to connect a monitor to each server. A common configuration today consists of a single keyboard, video monitor, and mouse that connect through a KVM switch to multiple servers. This arrangement saves both space and energy.
The Cumulative Effect
I've talked about a number of strategies that can reduce the amount of power used by any given computer. While the benefit for a single PC may be small, the aggregate effect among all the computers in an organization can yield measurable, even significant, savings. When you consider that some libraries manage dozens or hundreds of computers and that some university campuses house many thousands, reducing the energy consumed by all that equipment can have quite an effect. Imagine the impact if all CIL readers activated the powerreducing features in all of their organizations' PCs. It's a good way to help the environment and save a bit on library budgets at the same time.