Power consumption of datacenters poses economic, societal and environmental concerns. A large datacenter spends millions of dollars in yearly operational expenditures (op-ex) paying its electricity bills, and in provisioning the power delivery network, to accommodate the peak power draw, even if this draw is rarely sustained. Re-shaping the power draw temporally (over time) and spatially (across regions of the datacenter, or even across datacenters), broadly referred to as Demand Response (DR), is a common technique to reduce cap-ex and op-ex costs. Until now, datacenter DR has primarily used computing knobs -- consolidation, scheduling, migration, and power state modulation -- to manage the power draw.
This project will investigate a novel set of complementary techniques leveraging Energy Storage Devices (ESDs) in the datacenter for DR. The project will explore specific issues related to provisioning -- type, capacity and location -- ESDs in the datacenter power hierarchy. It will also develop control techniques for sourcing from these devices, and coupling their usage with current computing DR techniques, all the while optimizing for performance, cost, availability and ESD health related criteria. This research will have a direct impact on datacenter power cost reduction and the ability to source renewable energy. The platforms, models and power traces from this work can be useful to further ESD research in datacenters. This project will also enhance interdisciplinary curriculum in the related areas, and lead to new undergraduate honor's thesis projects.
