Joint data center cooling and workload management: A thermal-aware approach

• Published in: Future generation computer systems Vol. 104; pp. 174 - 186
• Main Authors: MirhoseiniNejad, SeyedMorteza, Moazamigoodarzi, Hosein, Badawy, Ghada, Down, Douglas G.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2020
• Subjects: Cooling unit control; Data center power efficiency; Data center workload assignment; Efficient cooling; Thermal model; Thermal-aware scheduling; Data center power efficiency; Cooling unit control; Thermal-aware scheduling; Data center workload assignment; Efficient cooling; Thermal model
• ISSN: 0167-739X; 1872-7115
• DOI: 10.1016/j.future.2019.10.040

Information technology (IT) equipment and cooling infrastructure are key contributors to the total energy expenditure in a data center. There is typically significant power wastage due to inefficient cooling control and thermal-oblivious management of workload. Recent thermal-aware data center management techniques have not taken a unified approach in controlling IT and cooling systems. In this paper, we find that considering thermal effects of server workloads, in conjunction with control parameters of the cooling unit saves more power than optimizing each of them separately. We leverage a low complexity holistic data center model that considers thermal interactions between IT and cooling unit entities. This thermal model provides control decisions with fine-grained control variables. We propose joint cooling and workload management (JCWM), which has the potential to save a considerable amount of power by exploring synergies between the workload scheduler and operational parameters of the cooling unit. In addition, we provide a significant caveat for the power efficiency of server consolidation methods when taking into account associated thermal effects. •Cooling unit settings are controlled to reduce server over-cooling.•Joint cooling unit control of cooling and workload assignment significantly reduces cooling power consumption.•Server consolidation can paradoxically increase cooling power consumption.•Determining optimal server consolidation configurations is explored.