Remember me

Energy efficiency in IT

Published on 6 July 2009

Search the magazine archive





By Chris Edwards

off/on symbol

PCs could be made more energy efficient with emerging technology, reports E&T.

The International Energy Agency recently issued a stark warning. The Paris-based body claimed worldwide growth in IT and gadgets would see energy demands from those devices treble by 2030 unless something was done. Even with best practice, the number of devices expected to be in use within 20 years would cause energy consumption to grow 1 per cent a year, calling for a 20GW increase in generating capacity.

For Alain Anglade of the French environmental agency Ademe, the reason for the colossal increase is simple: “White goods used to be the main consumers of non-specific electricity. Now it’s computers.” Why are we in this situation? Moore’s Law is a big driver in this aspect of energy consumption, Anglade says: “I have a personal translation of Moore’s Law: the smaller circuits become, the more they need in energy and raw materials. We can see that in life-cycle analyses and in products.”

So much of the energy that computers consume in a day is in vain. In recent years, data centres have come under scrutiny because they concentrate so much power in one place - but that concentration covers-up the real offenders.

Google it

Earlier this year, Alex Wissner-Gross, head of CO2Stats, claimed a search on Google used as much energy as boiling a kettle. Although attention focused on Google’s role, the reality was that the search engine itself constituted very little of the total. Most of the energy, which assumed a 15-minute search session rather than a quick tap on the ‘I feel lucky’ button, was down to the user’s own PC and display.

According to Bruce Nordman of the Energy Analysis Department at Lawrence Berkeley National Laboratory in California, the devices at the edge of the network consume more than those running the network or the data centres. Worse, most energy use occurs when no-one is even using the machine - and that is likely to get worse before it gets better.

PCs take a long time to boot-up and shut down, especially in enterprise environnments, so lots of users don’t power down. The display may go blank after 30 minutes or so, but the system unit remains on, and idley so; but, during that idle period the PC is consuming a large percentage of the power that it would need when running demanding applications.

Hardware has improved over the past ten years, but it’s not unusual for an idling PC to draw 70W. Multiply that by 16 hours - assuming that you’re using the machine for the other eight during a work day - and over the course of a month one machine could have burnt around 23kWh; and that assumes a PC gets switched off on a weekend.

When the PC ‘wakes up’, it does not necessarily do sensible things. As Nelson points out, a PC operating system such as Windows has a habit of trying to renew its IP address with the server when the network card is not actually able to do anything about it. It may then continue using an address that has already been issued to another machine in the meantime.

Sleeping computers

If you want to log into a PC remotely, it has to be idling, not ‘asleep’. You cannot even put the machine into the suspend-to-RAM state, which still chews through more electricity than the deep-sleep state. It has to sit there, waiting patiently for someone to log in.

To be sure, the fact that your PC cannot go to sleep and wake reliably is not your fault. It’s bad design. And it’s something that researchers have realised.

Some ten years ago, Ken Christensen of the University of South Florida came up with an answer: let a proxy do the job of waiting and listening on behalf of the much more power-hungry x86 processor. It has taken a while, but the concept will form part of a revised Energy Star specification for power-saving PCs that goes live in July.

Christensen’s idea was to have a server do the waiting on behalf of a collection of clients. Although this is still one way of implementing the technique, attention is now focusing on the network interface card inside the PC itself - the idea is to run a minimal operating system on a network processor that monitors connections.

Having developed the core ideas, Christensen got in touch with Nordman at LBL who took the idea to the Energy Star committee, and has been instrumental in setting up a working group at ECMA to thrash out a standard way of designing proxies. “It has been talked about in an Energy Star context for at least four or five years. Since it was clear that Ken and I could not get this technology into the product, we would have to get industry to do that. We arrived at a critical mass last year,” Nordman says.

Demonstrations of the tech--nique include a ‘green telnet’ put together by Christensen’s team that maintains a terminal session between a client and server even while the client sleeps. There is also an Internet relay chat system that caches responses and passes them onto the client when it wakes.

A group at Microsoft Laboratories has demonstrated a USB stick called Somniloquy that provides the proxy functions for existing PCs. The network environment, however, is complex, and the proxying system cannot rely on changes to existing protocols - it has to be invisible on the network - which adds complexity. “There are lots of details and edge conditions; and then there are issues internal to the PC,” Nordman explains. “Typically, when a PC wakes up after a sleep, what the processor does is to reboot the network interface chip. If the network interface chip is supposed to remember things and you reboot it, that’s not so good,” Nordman explains.

A big question for the proxying technique is how quickly it will arrive. Although the new Energy Star specification cites it, there are no systems about to hit production that use it. The big chipset makers and operating system providers are represented on the ECMA working group but it could take some time before a new option to use proxying appears in a Windows or OS X power-management dialogue.

Nordman is convinced it is coming, as the system is relatively cheap to implement: “It is more a question of when rather than if. In the future why would we ever turn the PC off? Turning it off is a very 20th century idea. The future is sleep, not off. People will assume that the device is always connected to the network. Anything less would be stupid.”

Further information:



Comments

All comments

You need to be registered with the IET to leave a comment. Please log in or register as a new user.

Toolbox

Comment on this article

  • Case study

    Liverpool’s campus claw-back

    Lisa Nelson, principal programmer and analyst at the University of Liverpool, was stunned by how much power the PCs there had been wasting before she had her power-saving software installed on them in a rollout that started in May 2006. Up to that point, most of the ‘walk up’ PCs used by students were left running 24-hours a day, as were many staff machines. The PowerDown software that Nelson wrote would turn off each machine if left unattended with no-one logged in after around 30 minutes. In the space of one month, with the software running on close to 4,000 machines, she used system logs to calculate that some one million hours of running time were saved.

    The university is now rolling-out new software to a wider base of users, this time using the PowerMan application written by Data Synergy. “It does the pieces that Microsoft ought really to have written,” reckons Nelson.

    Nelson found when putting together her own control software how bizarre some of the decisions made by operating system creators are when it comes to power management. Power settings move around with the user, not with the machine, in Windows. And when the last user logs off, the machine runs with the power settings of that last user.