Workload management for air-cooled data centers: An energy and exergy based approach. (15th October 2020)

• Record Type: Journal Article
• Authors: Gupta, Rohit; Moazamigoodarzi, Hosein; MirhoseiniNejad, SeyedMorteza; Down, Douglas G.; Puri, Ishwar K.
• Is Part Of: Energy. Volume 209(2020)
• Publication Date: 2020-10-15
• Subjects: Data center -- Workload management -- Energy analysis -- Exergy analysis -- Irreversibility -- Multi-objective optimization; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.118485
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: The energy required to cool an air-cooled data center (DC) contributes significantly to the cost of operation, which is further exacerbated due to a poor choice of cooling architecture and ineffective IT workload management. Although existing algorithms reduce energy consumption, they do not minimize thermodynamic irreversibility by design. We provide a tradeoff approach that simultaneously minimizes power usage effectiveness P U E and maximizes the exergy efficiency η 2 n d . The temperature field is predicted inside a contained single-rack DC that is equipped with a rack-mountable cooling unit (RMCU) based on a mechanical resistance model for the fluid flow. This thermal model informs a multi-objective optimization framework based on a genetic algorithm to determine the optimal decision variables and tradeoffs for P U E and η 2 n d . We investigate the interrelated effects of (1) guidelines that ensure the reliability of the IT equipment, (2) overall network traffic load, (3) spatial IT load distribution, (4) changes in cooling system variables, and (5) multi-objective optimization. Results for the single rack system are presented in a scalable dimensionless form that is applicable for a multi-rack DC containing RMCUs. By considering the first and second laws of thermodynamics, this novel approach improves workload scheduling from both energy and exergy perspectives. Highlights: An approach to assign the IT load in a single-rack data center is provided. Joint use of workload and cooling variables decreases energy use and irreversibility. The dimensionless number θ × β and IT load distribution regulate spatial overcooling. Tradeoffs are obtained by solving single and multi-objective optimization. Considered together, variations in P U E and η 2 n d provide the basis for decisions.