An article in a popular business magazine published in 1999 claimed that by 2009, half of all the electricity consumed in the U.S. would go to power the Internet. That claim never came close to being true. In fact, electronics of all sorts today accounts for about 1% of the total demand from utilities. Nevertheless, that 1% comes to roughly 290 terawatts of electricity, a figure that should get our attention, says Lawrence Berkeley National Lab researcher Bruce Nordman. Speaking at the recent Berkeley Symposium on Energy Efficient Electronic Systems, Nordman described the current thinking aimed at dialing down the power consumed by electronic products in offices and homes.
Nordman claims that a lot of insights about how to reduce power dissipation come from examining data network usage. Typically networks are lightly used and they will almost certainly remain that way in the future. In fact, the average network operates at about 2% of its maximum capacity in most day to day chores. That opens up an opportunity to power down network resources when they aren't active.
Another insight comes from how the average office PC is used. It turns out that most are active only about 2% of the time. So the PC could be asleep rather than just idle when its operator is away. “There is a small benefit from convincing people to switch their machine from on to off, but there is a huge benefit available from convincing them to go from on to asleep,” Nordman says.
He also thinks usage statistics for consumer items such as TVs and set-top boxes are probably similar, and so are the potential savings from sleep modes.
These insights can be teamed with what we know about energy dissipation in electronics to attack unnecessary usage, Nordman says. For example, networks such as Ethernet are typically built to handle their maximum capacity. And many of the assumptions about how to build them are tied to considerations for minimizing latency. But for energy conservation, it is better to consider the average conditions. Of course, this is only possible if there are statistical profiles available of how the network resources are utilized.
In most cases it is surprisingly easy to assembly such statistics, says Nordman. They can be a source of opportunity for exposing ways that devices “at the network edge” can optimized for energy use. For example, routers at LBNL know how many devices connect to them. So they also can detect when a PC gets turned off or is asleep. IT people at the facility used such information to ascertain that three quarters of all PCs there are fully on overnight and over weekends.
As more devices get networked together, says Nordman, it will be possible to get even more useful information that can direct energy efficiency efforts. One example of a project in this area is the Energy Efficient Ethernet (EEE). To guide efforts at adding power management to Ethernet, it uses the idea that most links are idle most of the time. Researchers' initial approach to saving energy was to propose a way of switching between transmission speeds in a matter of milliseconds, much faster than the facilities in the existing specification. A better way emerged when an Intel researcher came up with the idea of having network nodes stop listening between packets. The switching can be accomplished in microseconds, says Nordman.
Look for EEE standards (dubbed IEEE 802.3az) to begin emerging in another year, says Nordman. The data centers typically dissipate about one watt per link per gigabit/sec of bandwidth, so the gains from EEE are potentially significant, he says.
Another power management standard under development will let network nodes such as PCs and set-top boxes export their presence on the network to some proxy device (perhaps a router) before going into a sleep mode. The proxy device would respond to routine network traffic in the name of the real device. The rest of the network perceives the device as still on. The upside: “We'll be able to save more than 50% of PCs' current consumption, a goal well worth getting,” says Nordman.
Finally, the biggest source of potential energy savings is in consumer electronics, says Nordman, though there is less of a distinction today between IT equipment and consumer electronics. The danger is rushing into standards in this area too quickly, particularly when considering standards for networks inside buildings. Efforts today in this area could easily lead to increased use of electricity rather than savings due to lack of foresight. Programmable thermostats are an example: Energy use goes up when they are installed rather than down because they are generally cumbersome to use, he says.
What is needed in buildings is not digitally networked versions of building controls, but something fundamentally new. Specifically, he advocates a goal of universal interoperability: Any energy using device can operate in any room or building on the planet and be used by any person. Unfortunately, standards reflecting this kind of goal are still in the future.