HPC Data Center Participation in Demand Response: an Adaptive Policy with QoS Assurance

Area of Science:

• Computer Science
• Electrical Engineering
• Operations Research

Background:

• Demand response programs incentivize grid stability by adjusting consumer power usage.
• Regulation service requires rapid power adjustments, aligning with HPC data center flexibility.
• Existing research on data centers in regulation services overlooks Quality-of-Service (QoS).

Purpose of the Study:

• To develop a data center policy for participating in regulation services while guaranteeing job QoS.
• To address the gap in prior work concerning QoS provisioning in data center demand response.

Main Methods:

• Proposed an Adaptive policy integrating job scheduling and server power capping for power regulation.
• Utilized queueing theory to ensure job QoS within the scheduling strategy.
• Conducted experiments on a real HPC cluster to validate the policy.

Main Results:

• The Adaptive policy reduced electricity costs by 25-56% compared to baseline policies.
• Guaranteed job QoS was maintained throughout the regulation service participation.
• Baseline policies failed to meet QoS constraints in a significant portion (9 out of 14) of tested workloads.

Conclusions:

• The proposed Adaptive policy effectively enables HPC data centers to participate in regulation services.
• The policy successfully balances grid stabilization objectives with essential job QoS requirements.
• This approach offers a practical solution for cost reduction and reliable operation in data centers engaged in demand response.