Modern cloud systems consume enormous amounts of energy. Large-scale datacenters power millions of servers continuously, and even small improvements in system efficiency can translate into significant reductions in operational cost, cooling requirements, and environmental impact. At the same time, the rapid growth of AI workloads, microservices, and edge computing has made energy consumption an increasingly important systems concern. Despite this, most software systems today are still optimized primarily for performance, with energy efficiency treated as an afterthought.
Aurora is a project that explores how systems can automatically adapt themselves to become more energy efficient at runtime. The central idea is to integrate fine-grained energy and power measurements into the Iridescent runtime so that specialization decisions can optimize not only for performance, but also for energy consumption and energy-delay tradeoffs. Concretely, the project will explore mechanisms for incorporating telemetry from power measurement interfaces into the specialization loop. This would allow the system to evaluate candidate specializations using energy-aware objective functions and dynamically select implementations that minimize energy usage for a given workload profile and deployment setting. We believe this direction could be particularly valuable for modern cloud and edge systems where energy efficiency is increasingly becoming a first-order systems concern.