[Paper Review] Incentivizing Users of Data Centers Participate in The Demand Response Programs via Time-Varying Monetary Rewards

Yong Zhan, Xu Du+2 more|arXiv (Cornell University)|Apr 7, 2016

Smart Grid Energy Management19 references3 citations

TL;DR

This paper proposes a time-varying monetary reward mechanism to incentivize data center users with flexible deadlines to voluntarily defer their workloads, enabling demand response without performance penalties. Using a game-theoretic framework, the approach ensures non-intrusiveness, fairness, and cost guarantees, reducing peak load by up to 20.9% and overall electricity costs by 7.4% in real-world traces.

ABSTRACT

Demand response is widely employed by today's data centers to reduce energy consumption in response to the increasing of electricity cost. To incentivize users of data centers participate in the demand response programs, i.e., breaking the "split incentive" hurdle, some prior researches propose market-based mechanisms such as dynamic pricing and static monetary rewards. However, these mechanisms are either intrusive or unfair. In this paper, we use time-varying rewards to incentivize users, who have flexible deadlines and are willing to trading performance degradation for monetary rewards, grant time-shifting of their requests. With a game-theoretic framework, we model the game between a single data center and its users. Further, we extend our design via integrating it with two other emerging practical demand response strategies: server shutdown and local renewable energy generation. With real-world data traces, we show that a DC with our design can effectively shed its peak electricity load and overall electricity cost without reducing its profit, when comparing it with the current practice where no incentive mechanism is established.

Motivation & Objective

• To address the 'split incentive' problem where data centers and users lack aligned incentives for demand response participation.
• To design a non-intrusive, deadline-aware, and fair reward mechanism that allows users to opt-in voluntarily.
• To ensure users' maximum costs do not exceed baseline costs under the new pricing model.
• To integrate the reward mechanism with existing demand response strategies like server shutdown and renewable energy generation.
• To minimize the data center’s electricity cost while preserving its profit through optimal time-varying rewards.

Proposed method

• A game-theoretic model is formulated between a single data center and its users, capturing strategic decisions on workload deferral.
• Users' surplus is modeled as a function of performance degradation and time-varying rewards, with dominant strategies derived based on reward thresholds.
• The data center’s cost minimization problem is formulated as a convex optimization program, with optimal time-varying rewards derived as solutions.
• The reward system ensures fairness by paying only users who defer requests, with rewards proportional to their contribution.
• The mechanism is extended by integrating with server shutdown and local renewable energy generation strategies to enhance overall demand response effectiveness.
• Baseline costs are preserved via a max-cost guarantee, ensuring users are not worse off than under flat-rate or usage-based pricing.

Experimental results

Research questions

• RQ1 How can a data center incentivize users with flexible deadlines to voluntarily defer their workloads without imposing performance penalties?
• RQ2 What time-varying reward structure ensures fairness, non-intrusiveness, and user cost guarantees?
• RQ3 How does the game-theoretic interaction between the data center and users affect workload deferral decisions?
• RQ4 What is the optimal reward policy that minimizes the data center’s electricity cost while preserving profit?
• RQ5 How does integrating the reward mechanism with server shutdown or renewable energy generation improve demand response outcomes?

Key findings

• With real-world YouTube U.S. data traces, the proposed mechanism reduced the data center’s peak electricity load by 20.9%.
• The overall electricity cost was reduced by 7.4% compared to baseline practices without incentives.
• The mechanism maintained user cost guarantees, ensuring no user paid more than under traditional pricing models.
• The integration of time-varying rewards with server shutdown further reduced peak load and electricity costs.
• The integration with local renewable energy generation also led to additional reductions in peak load and overall electricity costs.
• Theoretical analysis confirmed that users adopt dominant strategies based on reward thresholds, validating the mechanism’s strategic robustness.