Aggregation-Based Datacenter Energy Management in Wholesale Electricity Markets

Miao PanarXiv 2017 · arXiv

TL;DR

This paper proposes a cooperative game theory approach for datacenters to collectively bid in wholesale electricity markets, reducing their overall energy costs by mitigating demand uncertainty. They introduce a novel cost allocation scheme that ensures fair distribution of savings and stability within the cooperative.

What Problem Does This Paper Address?

Datacenters are massive energy consumers, with electricity costs forming a significant portion of their operating expenses. While individual datacenters can participate in wholesale electricity markets to potentially lower costs compared to retail, this exposes them to risks from uncertain power demand and fluctuating market prices. Deviations from day-ahead commitments lead to penalties in the real-time market. This paper addresses how multiple independent datacenters can overcome these individual risks by forming a coalition to bid collectively, thereby reducing overall demand uncertainty and its associated penalties.

Methods

The core of the paper lies in modeling the datacenter aggregation as a cooperative game with transferable cost.

  • Cooperative Game Formulation: Datacenters are treated as players in a game. The cost function for any coalition (group of datacenters) is defined as the minimum expected energy cost they can achieve by collectively bidding in the day-ahead market.
  • Cost Allocation:
    • Expected Cost Allocation: A novel marginal cost allocation scheme is proposed. This scheme distributes the total expected cost based on each datacenter's marginal contribution to the aggregated demand, ensuring the allocation lies within the "core" of the game (meaning no datacenter has an incentive to leave). This is computationally efficient, requiring only O(N) calculations.
  • Realized Cost Allocation: To address day-to-day cost variations, a proportional allocation method is introduced. This method distributes the actual realized cost based on the expected cost allocation, ensuring that over the long run, the average realized cost approaches the expected cost, thus maintaining coalition stability.
  • Modeling: A detailed datacenter power model is presented, incorporating IT power (based on CPU utilization and server count) and non-IT power (via PUE). Renewable energy generation (wind turbines) and a two-settlement electricity market (day-ahead and real-time) are also modeled.

Datasets

  • Workload Data: Google cluster trace [47] was used to simulate workload arrival rates over 29 days, extended to 1008 hours (42 days).
  • Renewable Energy Data: NREL National Wind Technology Center (M2) [48] dataset for wind speed was used to simulate wind power generation.

Key Insights and Ideas

  • Aggregation Reduces Uncertainty: The fundamental insight is that by pooling demand from multiple datacenters, the statistical diversity of their individual workloads and renewable generation smooths out the aggregated demand, significantly reducing the uncertainty faced by any single entity.
  • Cooperative Game Theory is a Natural Fit: Cooperative game theory provides a robust framework to model the strategic interactions and benefit sharing among independent datacenters aiming for a common goal (cost reduction).
  • Marginal Contribution for Fair Allocation: The proposed cost allocation based on marginal contribution ensures that each datacenter contributes proportionally to the overall cost reduction, making the scheme fair and incentivizing participation.
  • Realized Cost Allocation for Stability: The distinction between expected and realized costs is crucial. The proportional realized cost allocation mechanism is key to preventing datacenters from leaving the coalition due to short-term cost fluctuations.

Key Results

  • Significant Cost Savings: The proposed cooperative bidding strategy resulted in an average hourly cost saving of approximately 18.03% compared to individual bidding.
  • Core Allocation Achieved: The proposed cost allocation scheme was shown to lie within the core of the cooperative game, satisfying both efficiency and subgroup rationality.
  • Efficiency of Allocation: The marginal cost allocation method is computationally efficient, requiring O(N) calculations, which is a significant advantage over methods like the nucleolus.
  • Impact of Percentile: The study showed that the percentile choice for day-ahead bidding significantly impacts cost savings and bidding levels, with higher percentiles generally leading to higher savings but also higher bids.

Limitations and Future Work

  • Game is Nonconvex: The paper demonstrates that the cooperative bidding game is nonconvex, meaning standard methods like the Shapley value are not guaranteed to produce a core allocation. This necessitates the development of specialized allocation schemes.
  • Assumptions: The model assumes datacenters are price-takers in the market and that market prices are statistically independent of workload and generation.
  • Future Work: The paper suggests exploring more complex market structures and considering scenarios with more diverse datacenter characteristics and renewable energy sources. Further investigation into the impact of different percentile choices and their trade-offs would also be valuable.

Summary updated

Abstract

In this paper, we study how datacenter energy cost can be effectively reduced in the wholesale electricity market via cooperative power procurement. Intuitively, by aggregating workloads and renewables across a group of datacenters, the overall power demand uncertainty of datacenters can be reduced, resulting in less chance of being penalized when participating in the wholesale electricity market. We use cooperative game theory to model the cooperative electricity procurement process of datacenters as a cooperative game, and show the cost saving benefits of aggregation. Then, a cost allocation scheme based on the marginal contribution of each datacenter to the total expected cost is proposed to distribute the aggregation benefits among the participating datacenters. Besides, we propose proportional cost allocation scheme to distribute the aggregation benefits among the participating datacenters after realizations of power demand and market prices. Finally, numerical experiments based on real-world traces are conducted to illustrate the benefits of aggregation compared to noncooperative power procurement.
BibTeX
@misc{pan2017aggregation,
  author = {Miao Pan},
  title = {{Aggregation-Based Datacenter Energy Management in Wholesale Electricity Markets}},
  year = {2017},
  eprint = {1707.02631},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1707.02631},
}

Loading related content...

Figure