Rack Level Study of Hybrid Liquid/Air Cooled Servers: The Impact of Flow Distribution and Pumping Configuration on Central Processing Units Temperature

Abstract

The flow distribution and central processing unit (CPU) temperatures inside a rack of thirty 1U (single rack unit) Sun Fire V20z servers retrofitted with direct to chip liquid cooling and two coolant pumping configuration scenarios (central and distributed) are investigated using the EPANET open source network flow software. The results revealed that the servers in the top of the rack and close to the cooling distribution unit can receive up 30% higher flow rate than the servers in the bottom of the rack, depending on the pumping scenario. This results in a variation in the CPU temperatures depending on the position in the rack. Optimisation analysis is carried out and shows that increasing the flow distribution manifold's dimensions can reduce the flow variation through the servers and increase the total coolant flow rate in the rack by roughly 10%. In addition, activating the small pumps in the direct to chip liquid cooling loops inside the servers (distributed pumping) resulted in an increase of 2 in the CPU temperatures at the high computational workload.

Performance Comparison between Data Centers with Different Airflow Management Technologies

“…To calculate the energy consumption, the refrigeration system performance is required [40], as shown in Equation (8).…”

“…To reduce energy consumption, many efforts have been dedicated to study the performance of new technologies, such as liquid [6], two-phase [7], and integrated cooling systems [8]. Although these new technologies could largely improve the cooling efficiency and reduce energy consumption, the capital, operational, and maintenance costs as well as the payback period need to be further considered.…”