Digital fingerprinting is used in several domains to identify and track variable activities and processes. In this paper, we propose a novel approach to categorize and recognize computational tasks based on thermal system information. The concept focuses on all kinds of data center environments to control required cooling capacity dynamically. The concept monitors basic thermal sensor data from each server and chassis entity. The respective, characteristic curves are merged with additional general system information, such as CPU load behavior, memory usage, and I/O characteristics. This results in two-dimensional thermal fingerprints, which are unique and achievable. The fingerprints are used as input for an adaptive, pre-active air-conditioning control system. This allows a precise estimation of the data center health status. First test cases and reference scenarios clarify a huge potential for energy savings without any negative aspects regarding health status or durability. In consequence, we provide a cost-efficient, light-weight, and flexible solution to optimize the energy-efficiency for a huge number of existing, conventional data center environments.
Vodel, Matthias and Ritter, Marc, "Thermal Fingerprinting—Multi-Dimensional Analysis of Computational Loads" (2017). CONF-IRM 2017 Proceedings. 35.