Computing and Networks
As computing and data storage increasingly become a globally-available, public utility, the proliferation of large numbers of massive data ` centers will have a substantial energy footprint in our future. U.S. businesses now spend $3.3 billion to power their data centers, and that cost is growing rapidly. Energy expenditures are already becoming more significant than the cost of machines, making energy efficiency a critical factor in the future success of our nation’s IT industry. The Institute for Energy Efficiency brings a distinctively cross-disciplinary approach to designing energy-efficient data centers, applying world-class research in large-scale computing systems, low-power electronics and devices, and modeling and control of buildings. Network research in this area is about powering down access points in wireless networks when those devices are not in use (i.e. during the night), while still maintaining bare bones coverage in case a user does come into the network. As the load on the network increases, ‘on’ access points can wake up other, sleeping access points, to help handle the traffic load. In testbed network deployment, colleagues observed up to 45% energy savings with their solution.
Trend in power consumption of integrated circuits, showing the need for innovation, including 3D chips, multicore approaches and optical interconnects.
R. Schmidt, T.C. Chen, IBM
Within the Institute for Energy Efficiency is the GreenScale Datacenter Initiative, which serves as a focal point for a vertically-integrated effort to drive key energy-efficient technologies in computing, electronics, and building systems. Collectively, these technologies address very significant near-term and long-term energy challenges. GreenScale provides a focused testbed in which to develop synergistic approaches to managing massive computation and data storage, acquiring and assimilating data from networks of sensor and control systems, developing components and technologies for low-power computing and data storage, and modeling and control of building systems.
The hardware problems are twofold. One is a rapid increase in the number of servers. Google has 1 million servers and is adding 400,000 servers per year. This drives research into more efficient processors. The second problem is power density as chips get faster and more dense. UCSB is leading research into optical interconnects to reduce the power required for interconnecting arrays of processors.
