Skip to main navigation Skip to search Skip to main content

Optimising Thermal Performance in Data Centre Server Racks via a Parametric Layout Configuration Study

  • Xinkun Gao
  • , Mingqin Cai
  • , Ruiyong Mao
  • , Cheng Hu
  • , Hongwei Wu
  • , Qiaoxinyi Zhou
  • , Xiaoyan Yi
  • , Zujing Zhang
  • , Weiping Yang

Research output: Contribution to journalArticlepeer-review

Abstract

The rapid expansion of data centres has intensified the demand for low-threshold thermal optimisation strategies that avoid the high costs and operational disruptions associated with conventional infrastructure retrofits. This study focuses exclusively on in-rack server layout adjustments, specifically server power combination, arrangement methods, and air supply velocity. A dedicated experimental platform was constructed to represent a standard server rack environment, and physical single-factor and orthogonal experiments were conducted to evaluate cooling efficiency. Results indicate that: (1) Heat accumulates primarily at the rack top, producing pronounced temperature stratification. (2) The relative impact of the examined parameters on thermal performance ranks as follows: air supply velocity > power combination > arrangement method. (3) An optimal configuration—400 W × 10 servers positioned in the lower rack with an air supply velocity of 4 m/s—effectively alleviates heat imbalance and suppresses hot-spots. Compared with a non-optimised baseline, this configuration yielded measurable improvements in airflow utilisation and thermal uniformity. This work demonstrates that strategic in-rack layout adjustments alone can markedly enhance cooling performance without additional capital investment or infrastructure modifications, offering actionable guidance for data centre operators seeking practical, low-effort optimisation measures.
Original languageEnglish
JournalScientific Reports
Publication statusAccepted/In press - 16 Jun 2026

Fingerprint

Dive into the research topics of 'Optimising Thermal Performance in Data Centre Server Racks via a Parametric Layout Configuration Study'. Together they form a unique fingerprint.

Cite this