Energy-Proportional Computing and Climate Change

The importance of doing nothing well will play a big role in the conservation of energy and the fight against climate change.

According to a 2007 study by the Institute of Electrical & Electronics Engineers, an energy-efficient computer server consumes 50 percent of its peak power when idle. The article pushed for energy-proportional computing, in other words, to consume more power as you compute more.

This may sound intuitive, but it is not how designers of many computers and more importantly, computer networks implement their systems today. The relevance to climate change becomes apparent when one considers that computers contribute the same amount of greenhouse gas emissions as aviation according to a report published by the Climate Group, and overall percentage from computers will grow by 2020 if business as usual continues.Engineers will have to undertake both monotonous and revolutionary projects to arrive at energy-proportional computing. The largest consumers of energy in the IT space (e.g., Google and Microsoft) are working to solve this problem at brisk pace, but even Google’s servers sit idly 30 to 40 percent of the time, according to Professor Randy Katz of the University California at Berkeley. A group he is spearheading at Berkeley called LoCal is attempting to define and implement the steps to progress this herculean effort.

To “do nothing well,” LoCal says the following must happen:

• Engineers must architect networks that wake up and go to sleep faster.
• Network designers must challenge the “always on” assumption for desktops and appliances.
• Engineers must look beyond the computer network and consider the operation of the building housing the system and the electricity grid itself.

The first goal is a typical design problem for engineers: make the system faster and more efficient. However, accomplishing this goal will require designers to go beyond the typical constraints under which they operate. Networks will require significant improvements in scheduling and forecasting of work to allow more machines to go to sleep at any given moment.

The second goal builds off the first and is a challenge to a fundamental principle that exists in many networks today. A dynamic network capable of adapting to a changing topology and one that consumes energy proportional to the current demand is necessary.

The last goal highlights an opportunity rather than a problem for computer networks. LoCal is looking at the Internet as a model for the U.S. energy infrastructure. They incorporate many new energy concepts in their work including distributed generation, the smart grid, and demand response.

Unleashing computer scientists on the energy infrastructure in the United States could yield swift changes in how energy is distributed and consumed. Developments in this field could induce the changes in personal habits related to energy consumption that policymakers have struggled with for decades.