Causality-based Thermal Prediction for Data Center

Changes in operating conditions in a data center lead to heating-up of IT equipment and generation of hot-spots. A required supply of cold air at a specific temperature with an adequate flow rate needs to be maintained to avoid such hot-spots. To do so, appropriately identified Air Control Units (ACUs) that have the best causal effect on the cabinets where the IT equipment runs, need to change the air supply characteristics (i.e., supply-air temperature and flow rate). Hence, it is essential to determine such a relationship between changes in supply from an ACU to its effect on cabinets. In this paper, we study the use of Granger Causality to identify such a relationship between air supply sources and cabinets in a data center. Using an industrial data center simulator (6SigmaRoom), we show that the ACUs exhibiting Granger Causality with cabinets have a time-series inference based unidirectional temporal precedence relationship between them that establishes the relationship between the ACUs and cabinets.