|
|
 |
» |
|
|
|
|
Across the computing spectrum - from servers and supercomputers to personal computers and cell phones - power consumption is rapidly emerging as one of the key limiting factors impeding greater adoption of more advanced computing solutions.
| |
|
For consumers, the improved battery life from our technologies dramatically enhances the usability of mobile devices. Indeed, studies show that mobile users often rate the need for longer battery life higher than even the need for more performance. Power management also impacts the broader system functionality (e.g., form factor, weight and packaging) and user experience (e.g., multiple power adapters, feature set) of these devices.
For enterprises, the capital and recurring costs associated with power delivery, electricity consumption and heat extraction are dramatically increasing. Additionally, better power and heat management allows for more consolidation and reliability in future data centers. This can lead to more cost-effective enterprise backbones, enabling more widespread deployment of computing services, even for the cost structures of emerging markets like India and China.
Finally, power management has broader societal impact with its environmental benefits. For example, the energy consumed by compute equipment contributes to more than 4 million tons of carbon dioxide emissions annually.
Effective power management can address this dramatically. One indication of this is the importance environmental agencies worldwide (TopRunner, Energy Star) are giving to more aggressive standards to drive energy-efficient system designs.
Our research seeks to develop fundamental technologies for energy-efficient, high-performance computing at all levels of the system - from chips to data centers. Some key recent contributions include:
- Energy-adaptive displays and energy-aware user interfaces that pioneered the notion of displays that adapt their energy consumption based on scope of user interest; as a result, the display battery life improves two-fold to twenty-fold.
- Heterogeneous multi-core architectures that design core diversity into chip multiprocessors to better match workload requirements to architectural efficiency; this design enables two-fold to ten-fold improvements in power.
- Ensemble-level power management for future blade servers that enforce the power budget in software, and holistically across a collection of servers (e.g., at the enclosure level); for current enterprise deployments, this can improve the power by a factor of two.
- Facilities-aware data center resource provisioning that adapts workload scheduling to optimize for power and cooling costs in addition to performance; for example, temperature-aware resource scheduling can move heat-generating workloads to cooler locations of the data center and reduce cooling costs by half.
Our research has also resulted in several new approaches for power measuring and monitoring - including Joulesort, energy scale down, energy-based statistical profiling, location-aware knowledge planes, as well as proxy-based environmental modeling.
Contact Partha Ranganathan for more details.
|
Printable version
