Modeling Thermal Mass of a Data Center Validated With Actual Data due to Chiller Failure

Abstract

The task of minimizing the downtime of a data center is becoming increasingly important due to the necessity of availability and maintaining the integrity of the data being handled by the data center. Consequently, a model used to predict the thermal response of a data center would be useful information in designing mechanisms to minimize the downtime during a failure or to serve as an alternative analysis method other than CFD. This paper will focus on a thermodynamic approach of predicting the thermal response of the data center space with the use of lumped system analysis. The model will be developed and validated using actual data from a chiller failure event in the CEETHERM Data Center Laboratory. Events in sequence are: (i) Chiller failure, (ii) Data center shutdown due to critical temperatures and (iii) Chiller restored. To illustrate, the data center section of interest consists of 10 racks of servers (maximum capacity of 24kW for each rack) with a total of 3360 nodes and is chilled using chilled water from the building chiller, through which the cooling resources are distributed using a rear door heat exchanger and a cooling room air conditioning unit (CRAC). The relevant and important data that was recorded in this failure are the: (1) Server inlet temperatures, (2) CPU temperatures, (3) CRAC supply and return air temperatures, (4) Chiller supply and return water temperature, (5) Chiller flow rate, (6) Data center space temperature and humidity, (7) Server power draw and (8) CRAC fan speeds.