Session: 10-01 Interactive Presentations

Paper Number: 99128

99128 - Feasibility Study of Rear Door Heat Exchanger for a High Capacity Data Center 

Due to increased use of high-performance computing in datacenters, old low-performance compute servers must be replaced endlessly with high-performance compute servers. Traditional air-cooling systems are insufficient or energy inefficient to provision the servers in optimal conditions as the datacenter thermal footprint or rack density grows. To sustain the growing needs, Rear Door Heat Exchangers (RDHx) are deployed in existing datacenters. RDHx are typically built to fit at the rear end of the rack which is 2ft wide. A commercially available rear door heat exchanger was chosen to fit at the back of the ORV2 rack from Open Compute Project (OCP). The rack consists of a 4-OU IT equipment, power shelves and a TOR switch. All the ITEs are provided with power ratio curves based on a 12°C ΔT. For the current study, the coolant flowrate for the RDHx is controlled using an internal controller that is modulated with respect to the air temperature exiting the RDHx. If the exit temperature increases from the set-point 30°C, coolant flow rate increases until the requirement is met, or maximum flowrate is achieved.

For a 7 kW rack the corresponding air flowrate requirement is 1050 CFM, similarly, for 9 kW and 11 kW the air flow requirements are 1350 and 1670 respectively. The server fans are selected such that when RDHx is in place, the server fans must overcome the pressure drop across the Hx to push the air through it with minimal to no recirculation. The poster will also show the feasibility of designing a datacenter with only RDHx and no peripheral CRAC/CRAH units while maintaining the thermal envelop. From the study it could be observed that for the first two cases with 7 kW and 9 kW rack powers, the server fans were able to push air through the RDHx and cool it down below 30°C while maintaining minimum pressure in the cabinet with minimal recirculation. In the third case of 11 kW rack power, the server fans were not able to supply adequate air through the RDHx since the pressure inside the cabinet increased resulting in recirculation inside the cabinet. When RDHx was simulated with active mode using counter rotating fans for 11 kW, the pressure inside the cabinet reduced as a result of excess air flow through the RDHx thereby reducing the exit air temperature below 30°C.

An active RDHx is deployed in a typical datacenter layout with 20 kW ORV2 racks without traditional cooling units like peripheral CRAC/CRAH units or a penthouse style design with supply shafts based on evaporative cooling. The data hall considered is approximately 160’ x 175’ x 22’ (L x W x H) with a typical hot aisle cold aisle layout. From the results, it can be observed that the RDHx can provide a room neutral solution thereby eliminating the need of any mechanical cooling units. It will still require a chiller but given the operating condition of 26°C facility water, there is potential to lower chiller usage based on favorable ambient conditions. About 3-4°C variation in processed air is observed from top to bottom since facility water connection is defined at the top of the coil. This could be further investigated for different coil designs; however, authors do believe that spatial variation in temperature is not a risk as the air gets mixed and average server inlet temperature is about 29.7°C. The datacenter is designed to meet ASHRAE Class A1 temperature limits and if the mean server inlet temperatures exceed 32°C the solution is considered failed.

Presenting Author: Satyam Saini The University of Texas at Arlington