The increased use of cloud and other large scale datacenter IT services and the associated power usage has put the spotlight on more energy-efficient datacenter management. In this paper, a simple model was developed to represent the heat rejection system and energy usage in a small DC setup. The model was then controlled by a reinforcement learning agent that handles both the load balancing of the IT workload, as well as cooling system setpoints. The main contribution is the holistic approach to datacenter control where both facility metrics, IT hardware metric and cloud service logs are used as inputs. The application of reinforcement learning in the proposed holistic setup is feasible and achieves results that outperform standard algorithms. The paper presents both the simplified DC model and the reinforcement learning agent in detail and discusses how this work can be extended towards a richer datacenter model. CCS CONCEPTS• Computer systems organization → Sensors and actuators; • Hardware → Enterprise level and data centers power issues; Temperature control; • Computing methodologies → Reinforcement learning; Modeling and simulation.
Simulation tools for thermal management of data centers help to improve layout of new builds or analyse thermal problems in existing data centers. The development of LBM on remote GPUs as an approach for such simulations is discussed making use of VirtualGL and prioritised multi-threaded implementations of an existing LBM code. The simulation is configured to model an existing and highly monitored test data center. Steady-state root mean square averages of measured and simulated temperatures are compared showing good agreement. The full capability of this simulation approach is demonstrated when comparing rack temperatures against a time varying workload, which employs time-dependent boundary conditions.
In this paper, we explore the use of Reinforcement Learning (RL) to improve the control of cooling equipment in Data Centers (DCs). DCs are inherently complex systems, and thus challenging to model from first principles. Machine learning offers a way to address this by instead training a model to capture the thermal dynamics of a DC. In RL, an agent learns to control a system through trial-and-error. However, for systems such as DCs, an interactive trial-and-error approach is not possible, and instead, a high-fidelity model is needed. In this paper, we develop a DC model using Computational Fluid Dynamics (CFD) based on the Lattice Boltzmann Method (LBM) Bhatnagar-Gross-Krook (BGK) algorithm. The model features transient boundary conditions for simulating the DC room, heat-generating servers, and Computer Room Air Handlers (CRAHs) as well as rejection components outside the server room such as heat exchangers, compressors, and dry coolers. This model is used to train a RL agent to control the cooling equipment. Evaluations show that the RL agent can outperform traditional controllers and also can adapt to changes in the environment, such as equipment breaking down.
The purpose of this program was to design and develop a second-generation hydrofluidic servovalve and to demonstrate the performance of a servoactuator using the servovalve. The servovalve uses a hydrofluidic amplifier cascade input state that replaces the bellows-flapper-nozzle of conventional servovalves and a fluidic feedback transducer. The report has been reviewed by the Eustis Directorate, U. S. Army Air Mobility Research and Development Laboratory; it is published for the exchange of information and appropriate application. Mr. George W. Fosdick of the Systems Support Division served as project engineer for this effort. DISCLAIMERS The findings in this report are not to be construed as an official Department of the Army position uniess so designated by other authorized documents. When Government drawings, specifications, or other data are used for any purpose other then in connection with a definitely related Government procurement oaeration, the United States Government thereby incurs no responsibility nor any obligation whatsoever; and the fact that the Government may have formulated, furnished. or in any wy supplied the said drawings, specifications, or other date is not to be regarded by implication or otherwise as in any manner licensing the holder or any other person or corporation, or conveying any rights or permission, to manufacture, use, or sell any patented invention that may in any way be related thereto. Ttiilek iaiLs cigid it tis regt~t (to not constitute an blficlal endorsement or approval of the use of such 'ommecial hardwiare o software. DISPOSITION INSTRUCTIONS Destroy this report when no longer needed. Do not return it to the originator.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.