The growth of cloud platforms is raising significant concerns about their impact on the environment. To reduce their carbon footprint, future cloud platforms will need to broadly adopt low- or zero-carbon energy sources, such as solar or wind. A distinguishing characteristic of clean energy is its unreliability. Unfortunately, today's energy systems mask this unreliability in hardware, which prevents applications from optimizing their carbon-efficiency, i.e., work done per kilogram of carbon emitted. To address the problem, we design an ``ecovisor,'' which virtualizes the energy system and exposes software-defined control of it to applications. Our ecovisor enables each application to handle clean energy's unreliability within the software stack based on its specific requirements. We implement a small-scale ecovisor prototype that virtualizes a physical energy system capable of regulating power flow between the grid, a solar array, batteries, and a cluster of microservers. We evaluate our ecovisor's flexibility by showing how a range of applications can exercise their virtual energy system in different ways to optimize carbon-efficiency. For example, we show how using a 1kWh battery to reduce renewable energy volatility decreases the running time and energy usage of an elastic Spark job by 4.4x by eliminating recomputation overhead due to power shortages.
David Irwin is an Associate Professor in the Department of Electrical and Computer Engineering and an Adjunct Associate Professor in the College of Information and Computer Sciences at the University of Massachusetts Amherst. He leads the Sustainable Computing Lab, which focuses on designing distributed software systems with an emphasis on improving efficiency and sustainability.
