Reinforcement learning for energy improvements in data centers

• Online algorithm for workload and facility management to reduce energy usage through reinforcement learning.
• Demonstrates around 60% improved power usage efficiency (PUE) over the already world-class RISE ICE data center.
• The proposed holistic approach to datacenter management is cuttingedge.
• Publication: A. Heimerson, R. Brännvall, J. Sjölund, J. Eker, J. Gustafsson, ”Towards a Holistic Controller: Reinforcement Learning for Data Center Control”, 9th International Workshop on Energy-Efficient Data Centres (E2 DC 2021).

Input(s):

• Connected datacenter.

Main feature(s):

• Energy efficient control of cloud services using reinforcement learning.

Output(s):

• Algorithms

None

• Developers
• Researchers

Algorithms