Energy-Efficient Computing

• Applicant: Dr. Antonios Antoniadis
• Subject Area: Theoretical Computer Science
• Term: from 2016 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 317172119

The energy consumption of computing systems is increasingly becoming of critical importance. On one hand, data centers consume a big portion of the global electricity production and on the other hand, mobile devices are heavily constrained by their battery capacity. It is therefore not surprising that there has been a significant upsurge in the study of energy-efficient computing, and algorithms in particular, in recent years. This research project aims to conduct basic research on several challenges that arise in the area of energy-efficient computing and are algorithmic in nature. In particular we propose to conduct basic research in the two fields of (i) energy-efficient scheduling algorithms and (ii) near-threshold computing.Managing power directly at the processor level is the most obvious type of algorithmic problem that considers energy as a resource. The two most common power-management techniques are dynamic speed scaling and powerdown both of which emerge in the context of scheduling problems. Dynamic speed scaling refers to the ability of the processor to adapt its speed in order to balance its performance and energy consumption, and powerdown is about transitioning devices into a sleep state when not needed so as to ameliorate the time and energy cost of these transitions with the energy savings. We propose to develop and analyse algorithms for different multiprocessor settings equipped with these two power-management techniques, to analyze a specific scheme for combining speed-scaling with game theory that captures a common cloud-computing based scenario, and to improve the state of the art with respect to single-processor deadline-based speed scaling.We are also interested in scrutinizing energy consumption directly at the circuit level, by employing near-threshold-computing. In this setting, the circuit design must balance the conflicting demands of minimizing the energy used per gate, and minimizing the number of gates in the circuit. Our goal is to design circuits for simple relations that have provably minimal energy consumption, to look at how one sets the supply voltage(s) on a given circuit so that the circuit performs its computation reliably while using approximately minimal energy, and to formaly capture the class of relations/circuits in which heterogeneity is really worth employing.We expect that our results apart from extending the theoretical understanding of energy-efficient algorithms and circuit design, will also have practical impact.

• DFG Programme: Research Grants