As the computing needs of business and consumers continue to grow, mega data centers with tens of thousands of machines are being built the size of shopping malls that consume tens of megawatts of power. These data centers support e-commerce, financial services, health care, telecommunications, and a variety of other services critical to the society. Power and thermal management of servers in these data centers and the clients using them has become a critical issue from many perspectives including utility costs, environmental impact, space usage, etc. Although there is a substantial amount of ongoing research into these issues in both academia and industry, many of the approaches tend to be ad hoc and empirical in nature. The purpose of the workshop is to examine power and thermal issues at a much more fundamental level and thereby bring a scientific approach to designing, configuring and managing hardware and software in a way that comprehends not only the power/thermal issues but also the associated performance and reliability tradeoffs.
The broader impact of the workshop is expected to be in terms of guiding future research in this area that is important from technological, economic and sustainability perspective
The funds were used to support two workshops on the science of power managment. The rapidly increasing power consumption associated with information technology (IT) results in a myriad of adverse impacts including high utility costs, unsustainable thermal densities, poor space usage, and a substantial carbon footprint. In view of this, reducing IT power consumption has become a pressing priority at all levels of computer technology from semiconductor materials and processes all the way up to the design of entire data centers. Although there is already significant research and development activity around power management in academia and industry, power remains a very difficult subject to deal with in a scientific and formal way. A grand challenge for the scientific community is to understand the trade-offs between power consumption and computability at a much more fundamental level. Ideally one would like to determine the limits of computability under power/thermal constraints, design systems that approach these limits and quantify corresponding performance tradeoffs. The participants at the first workshop consisted of leaders in the practice and science of power management. The purpose of the workshop was to provoke discussion among experts, identify key research directions, and report key findings to NSF. The format of the workshop consisted of: Presentations from field experts on what the important power/thermal problems are now, and are likely to be in the foreseeable future at a variety of levels from logic design to control of large computing infrastructures; Brief presentations from scientists on the current state of power management science research, including discussion of emerging power/thermal issues that can benefit from a formal treatment; Break-out discussions about what a "Science of Power Management" might look like in the future, and which research directions offer the most promise of impact. After the meeting, the organizers and steering committee will wrote a white paper summarizing discussions, key findings, and next steps for the purpose of advising NSF about the future of the "Science of Power Management". The intention is that NSF would use the workshop findings to explore suitable mechanisms to support research in this important area. The second workshop was intended primarily for theoretical computer scientists interested laying the foundation for the fledgling science of power management. The workshop was held on the last day of the Green Computing Conference, and consisted of short presentations on potentially interesting research directions, followed by discussions. Slides for these presentations were posted on the web, and a survery of open problems was written by Marek Chroback and appeared in the newsletter of the Association of Computing Machinery Special Interest Group on Algorithms and Computation Theory.
