Thermal-Aware Hybrid Workload Management in a Green Datacenter towards Renewable Energy Utilization

Yuling Li, Xiaoying Wang, Peicong Luo, Qingyi PanEnergies 2019 · DOI

TL;DR

This paper tackles the massive energy consumption of datacenters by proposing a "thermal-aware hybrid workload management" strategy. It intelligently schedules flexible (batch) workloads to coincide with renewable energy availability and dynamically adjusts cooling temperatures, significantly boosting green energy utilization and reducing reliance on traditional power sources.

What Problem Does This Paper Address?

Datacenters are energy hogs, consuming a significant portion of global electricity and contributing to carbon emissions. While many are adopting renewable energy sources like solar and wind, their intermittent nature poses a challenge for consistent datacenter operations. This paper addresses the problem of effectively integrating these fluctuating renewable energy supplies into datacenter operations, specifically by managing both computing workloads and the energy-intensive cooling systems to maximize green energy use and minimize overall energy costs.

Methods

The core of the proposed solution is a "thermal-aware workload management" (TM) strategy. This method builds upon existing approaches by:

  • Hybrid Workload Consideration: It explicitly accounts for two types of workloads: critical interactive workloads (requiring immediate response) and batch-type workloads (which can be deferred).
  • Workload Shifting: Batch-type jobs are strategically shifted to times when renewable energy is abundant.
  • Renewable Energy Forecasting: An LSTM (Long Short-Term Memory) neural network is used to predict solar power generation, enabling more accurate scheduling decisions.
  • Thermal-Awareness: Crucially, the method dynamically adjusts the cooling system's supply air temperature. This involves "pre-cooling" during periods of surplus solar energy to prepare for future demand, thereby utilizing excess green energy and reducing the need for brown energy later. This also considers the relationship between cooling temperature and the Coefficient of Performance (CoP) of cooling devices.
  • Joint Optimization: The TM strategy jointly optimizes workload scheduling and cooling temperature to maximize renewable energy utilization and minimize overall power consumption (both IT and cooling).

The paper compares this TM strategy against three baseline methods:

  • Static Method (ST): Batch jobs are processed as soon as they arrive.
  • Load Balancing Distribution over Time (LB): Batch jobs are spread evenly across time slots.
  • Best Effort Strategy (BS): Batch jobs are scheduled based on predicted solar power, but without dynamic temperature adjustment.

Datasets

The paper utilizes publicly available solar power data from a sharing website for training the LSTM model and for simulation. Specific data points mentioned include solar power data for sunny and cloudy (rainy) days in February, March, and April 2018, used to forecast solar power for July 14, 2018, and May 8, 2018. The authors note that they did not use actual load tracking data (like Google traces) but instead randomly generated jobs to simulate workloads.

Key Insights and Ideas

  • The Synergy of Workload and Thermal Management: The most significant insight is that effectively managing datacenter energy requires a holistic approach. Simply shifting workloads isn't enough; dynamically adjusting cooling temperatures in conjunction with workload scheduling unlocks greater energy savings.
  • Pre-cooling as an Energy Buffer: Using surplus solar energy for pre-cooling is a clever way to "store" renewable energy's potential by reducing future cooling demand. This prevents wasting excess solar power and ensures capacity for when solar generation dips.
  • LSTM for Predictive Power: The use of LSTM for solar power forecasting is a practical application of advanced machine learning to address the inherent uncertainty of renewable energy sources.
  • Batch Jobs are Key Enablers: The flexibility of batch-type workloads is the primary enabler for this strategy, allowing for deferral and optimization around renewable energy availability.

Key Results

The TM strategy consistently outperformed the other methods across various scenarios:

  • Solar Utilization: On a sunny day, TM achieved 98.2% solar utilization, significantly higher than ST (66.2%), LB (66.8%), and BS (89.1%). This trend held for cloudy days as well, with TM reaching 99.5% utilization.
  • Power Consumption: Figures 5 and 6 show that TM minimized the use of brown energy, almost entirely relying on solar power when available. While TM's total power consumption was sometimes higher, this was due to its effective utilization of all available solar energy, including for pre-cooling.
  • Job Scheduling: TM demonstrated better job scheduling, executing more jobs during periods of sufficient solar power, especially after noon on sunny days, due to its pre-cooling capabilities.
  • Waiting Time: While ST had the lowest waiting time for batch jobs (as it processed them immediately), TM offered a good balance between energy saving and acceptable waiting times, particularly in SLA-constrained environments. On sunny days, TM's average waiting time was 4.8 hours, compared to ST's 0 hours but significantly better than LB (8.3 hours).

Limitations and Future Work

  • Performance Under Low Solar Conditions: The paper acknowledges that the TM strategy's effectiveness is somewhat reduced when solar energy supply is insufficient, particularly on cloudy or rainy days. While still outperforming BS and LB, its energy-saving advantage over ST diminishes in such scenarios.
  • SLA Constraints: The authors note that while TM is suitable for SLA-constrained environments, the energy-saving effect might not be as pronounced under very strict SLA requirements.
  • Future Work: The authors plan to extend their work by incorporating demand-response signals from the grid. This would allow datacenters to further optimize their load and power consumption by participating in grid-level response plans, further enhancing energy management.

Summary updated

Abstract

The increase in massive data processing and computing in datacenters in recent years has resulted in the problem of severe energy consumption, which also leads to a significant carbon footprint and a negative impact on the environment. A growing number of IT companies with operating datacenters are adopting renewable energy as part of their energy supply to offset the consumption of brown energy. In this paper, we focused on a green datacenter using hybrid energy supply, leveraged the time flexibility of workloads in the datacenter, and proposed a thermal-aware workload management method to maximize the utilization of renewable energy sources, considering the power consumption of both computing devices and cooling devices at the same time. The critical knob of our approach was workload shifting, which scheduled more delay-tolerant workloads and allocated resources in the datacenter according to the availability of renewable energy supply and the variation of cooling temperature. In order to evaluate the performance of the proposed method, we conducted simulation experiments using the Cloudsim-plus tool. The results demonstrated that the proposed method could effectively reduce the consumption of brown energy while maximizing the utilization of green energy.
BibTeX
@inproceedings{li2019thermal,
  author = {Yuling Li and Xiaoying Wang and Peicong Luo and Qingyi Pan},
  title = {{Thermal-Aware Hybrid Workload Management in a Green Datacenter towards Renewable Energy Utilization}},
  booktitle = {Energies},
  year = {2019},
  doi = {10.3390/en12081494},
}
Figure