Improving cooling efficiency of servers by replacing smaller chassis enclosed fans with larger rack-mount fans

Abstract

As a common practice in the data center industry, chassis fans are used to direct air flow independent from neighboring servers. However, these fans are less efficient compared to larger rack level counterparts and also operate at higher sound levels. In this study, a novel approach is proposed whereby the smaller chassis enclosed fans are replaced with an array of larger fans, installed behind the stacked servers that share air flow.As a baseline study for comparison of the current scenario, a CPU dominated 1.5U Open Compute server, with four 60mm fans installed within the server, is characterized experimentally for its flow impedance, air flow rate, effect on die temperature and power consumption for various compute utilization levels. Larger fans with a square frame size of 80mm are carefully selected and individually characterized for their air moving capacity and power consumption. CFD is used to simulate the system of stacked servers and larger fans to obtain its flow characteristics and operating points.The fan power consumption of the larger fans is determined experimentally at these operating points replicated in an air flow bench. Comparing with the base line experiments, this study predicts a significant decrease in fan power consumption, without conceding the flow rate and the static pressure requirements of the server.

Effectiveness of Rack-Level Fans—Part I: Energy Savings Through Consolidation

“…A previous study by the authors of this work showed computational fluid dynamics evidence based on empirical data that significant savings in fan energy are possible through consolidation of smaller, internal server fans to a larger rack-based fan configuration [8]. Additional details of this previous study included findings that a failure of a single fan out of the four in the server under study has the same impact on die temperatures, irrespective of the location of the failed fan.…”

“…A simple wooden frame and acrylic panel were constructed to hold the larger fan array at the back of the servers. A previous computational study by the authors indicated that a minimum distance of 20 mm was needed between the back of the servers and the fan array to ensure uniform distribution of airflow between all four servers [8]. Following this minimum requirement, the fan array for this study was placed 25.4 mm (1 in) from the rear of the servers.…”

“…An in-depth discussion of the selection process is provided in Ref. [8]. Detailed characterization of the fans was performed on an Air Flow Bench test chamber to generate the fan curves and confirm the expected efficiency increase with larger fans.…”

“…Details of this testing procedure can be found in Refs. [8] and [11]. Using the Air Flow Bench test chamber, at each operating point on the fan curve, the fan's total electrical power draw by a voltage direct current (VDC) power supply was recorded.…”

Effectiveness of Rack-Level Fans—Part II: Control Strategies and System Redundancy

“…Preliminary work [14] predicted savings of up to 55% in cooling power by replacing smaller, chassis enclosed (60 mm) fans with larger rack-mount units for a stack of four servers. The present studies (parts I and II) advance this work by experimentally validating maximum possible savings through deployment of 80 mm and 120 mm fans.…”