Colocation Demand Response: Joint Online Mechanisms for Individual Utility and Social Welfare Maximization

摘要

Data centers with high yet elastic energy demand are ideal candidates for participation in demand response programs. This paper studies emergency demand response (EDR) at multi-tenant colocation data centers (colocations). While the colocation has no direct control over tenants' servers, we design online mechanisms to incentivize and coordinate tenants' energy reduction. Our mechanism is online in nature, aiming to maximize not only social welfare but also tenant utility. Our main proposal is a truthful incentive auction that provides tenants with monetary remuneration for EDR energy reduction, minimizing social cost, which combines seamlessly with an online primal-dual framework for each tenant to schedule their delay-tolerant workloads. The online optimization at each tenant targets its utility maximization, concurrently reporting valuation functions for the tenant to participate in the auction. Our online algorithms achieve long-term performance guarantees in both tenants' utility and social welfare maximization, while fulfilling the EDR requirement with minimal diesel generation. We validate the efficiency of our algorithms through both the theoretical analysis and real-world trace-driven simulations.