Heating Homes with Servers: Workload Scheduling for Heat Reuse in Distributed Data Centers.

Area of Science:

• Computer Science
• Sustainable Energy
• Thermodynamics

Background:

• Data centers are significant energy consumers, generating substantial waste heat.
• Distributed data centers in residential settings offer a unique opportunity for heat reuse.

Purpose of the Study:

• To develop and validate a workload scheduling solution for distributed data centers that reuses server heat for residential heating.
• To optimize workload allocation for maintaining desired home temperatures using waste heat from IT equipment.

Main Methods:

• A constraint satisfaction model was developed for optimal workload scheduling.
• Two models were created to correlate heat demand with server workload: a thermodynamic model and a machine learning model (Gradient Boosting Regressor).
• The solution was validated using monitored data from an operational distributed data center.

Main Results:

• The machine learning model achieved a high correlation accuracy of 4.74% in predicting workload for desired ambient temperatures.
• The thermodynamic model achieved an 11.98% correlation accuracy for server heat and power demand.
• The proposed solution effectively distributed workload to meet temperature setpoints and align server power demand with heat demand.

Conclusions:

• The developed workload scheduling solution successfully integrates distributed data centers into residential heating systems.
• The study demonstrates the feasibility of reusing waste heat from servers to meet thermal comfort requirements, enhancing energy efficiency.