Optimal Workload Allocation for Distributed Edge Clouds With Renewable Energy and Battery Storage

Duong Thuy Anh Nguyen, Jiaming Cheng, Lele Wang

TL;DR

This paper proposes a holistic optimization framework for Edge Service Providers (ESPs) to manage distributed edge clouds powered by a mix of grid power, on-site renewables, and battery storage. By intelligently routing workloads and managing battery cycles, the system minimizes operational costs and carbon footprints while strictly adhering to latency and quality-of-service (QoS) requirements.

What Problem Does This Paper Address?

Edge computing brings processing power closer to users to reduce latency, but this distributed nature creates a massive energy management headache. Edge clouds (ECs) are often located in different geographic regions with vastly different electricity prices and carbon intensities (e.g., Wyoming has cheap power but a high carbon footprint compared to the national average).

Furthermore, renewable energy like solar and wind is intermittent. ESPs face a complex "spatial-temporal" puzzle: How do you decide which edge cloud should handle a specific user request at a specific time to minimize costs and emissions without violating the maximum delay a user can tolerate?

Methods

The authors formulate the problem as a Mixed-Integer Linear Programming (MILP) model. The framework optimizes three main pillars simultaneously:

  1. Workload Allocation: Deciding which Access Point (AP) sends requests to which EC based on propagation delay and server capacity.
  2. Energy Management: Modeling power consumption using Power Usage Effectiveness (PUE) and server utilization rates. It accounts for grid power, on-site renewable generation, and battery state-of-charge.
  3. Economic/Environmental Impact: The objective function minimizes a combination of unmet demand penalties, electricity procurement costs (minus revenue from selling excess energy back to the grid), and carbon taxes.

Key technical constraints include battery charging/discharging efficiency ($\eta$), grid capacity limits, and a "green" vs. "brown" workload balance.

Datasets

The study validates the model using a combination of real-world traces and geographic data:

  • Topology: An 8-node edge cloud network based on randomly selected Equinix EC locations.
  • Workload Traces: Real-world demand data from the GWA-DAS (Grid Workloads Archive).
  • Energy & Carbon Data: Electricity pricing and carbon emission factors sourced from the U.S. Energy Information Administration (EIA) 2021 reports.
  • Renewable Traces: Solar/wind generation data from the GWA-DAS repository.

Key Insights and Ideas

  • Energy Arbitrage: Batteries aren't just for backup; they act as a tool for "energy arbitrage." The system can store cheap grid power or surplus renewable energy during off-peak hours and discharge it when grid prices or carbon taxes spike.
  • The "Sell-Back" Advantage: The model includes a "sell-back" option where excess renewable energy can be sold back to the grid. The authors found that as the sell-back ratio ($\zeta$) increases, the incentive to over-provision renewable capacity grows, further lowering total operational costs.
  • QoS vs. Sustainability Trade-off: There is a clear tension between latency and green energy. Stricter delay thresholds ($D^{max}$) limit the ESP’s ability to route traffic to "greener" but more distant ECs, forcing the use of local, potentially "browner" grid power.
  • Elasticity Matters: The ratio of peak power to idle power in servers significantly impacts how much energy can be saved through intelligent workload distribution.

Key Results

  • Superiority of the Holistic Model: The proposed model (M0) consistently outperformed benchmarks that lacked either battery storage (M3) or the sell-back option (M2).
  • Cost Reduction: Increasing renewable energy availability (via a scaling factor $\Psi$) led to a near-linear decrease in total costs.
  • Battery Impact: In high-electricity-price scenarios, having a battery (M0 and M2) significantly buffered the ESP against price volatility compared to models without storage (M1 and M3).
  • Carbon Footprint: By incorporating carbon taxes directly into the MILP objective, the model naturally shifts workloads to regions with cleaner grids or higher renewable yields whenever the latency constraints allow.

Limitations and Future Work

While the model is robust, the authors acknowledge a few areas for expansion:

  • Deterministic vs. Stochastic: The current model assumes the ESP has perfect knowledge of future renewable generation and demand. Future work could incorporate stochastic optimization to handle the inherent uncertainty of weather and user behavior.
  • Flexible Scheduling: The current model processes workloads in the time period they arrive. The authors suggest investigating task deferral, where non-urgent background tasks could be delayed to coincide with peak renewable production.
  • Battery Degradation: The model uses a simplified linear battery dynamic. Real-world implementations might need to account for the long-term chemical degradation of batteries caused by frequent cycling.

Summary updated

Abstract

This paper studies an optimal workload allocation problem for a network of renewable energy-powered edge clouds that serve users located across various geographical areas. Specifically, each edge cloud is furnished with both an on-site renewable energy generation unit and a battery storage unit. Due to the discrepancy in electricity pricing and the diverse temporal-spatial characteristics of renewable energy generation, how to optimally allocate workload to different edge clouds to minimize the total operating cost while maximizing renewable energy utilization is a crucial and challenging problem. To this end, we introduce and formulate an optimization-based framework designed for Edge Service Providers (ESPs) with the overarching goal of simultaneously reducing energy costs and environmental impacts through the integration of renewable energy sources and battery storage systems, all while maintaining essential quality-of-service standards. Numerical results demonstrate the effectiveness of the proposed model and solution in maintaining service quality as well as reducing operational costs and emissions. Furthermore, the impacts of renewable energy generation and battery storage on optimal system operations are rigorously analyzed.
BibTeX
@inproceedings{nguyen2024optimal,
  author = {Duong Thuy Anh Nguyen and Jiaming Cheng and Lele Wang and Duong Tung Nguyen},
  title = {{Optimal Workload Allocation for Distributed Edge Clouds With Renewable Energy and Battery Storage}},
  booktitle = {Icnc},
  year = {2024},
  doi = {10.1109/icnc59896.2024.10555962},
}
Figure