A transient model for data center thermal prediction

Fast thermal maps are a crucial component for many green data center design techniques. However, most state of the art work on thermal mapping ignores critical temporal aspects of thermal behavior and relies on modeling assumptions, such as the steady state assumption, that can reduce their accuracy and cause heat-induced performance throttling when used for task scheduling. These problems have the potential to affect the energy savings projected by such models. This paper introduces a fast thermal modeling technique that captures the transient behavior necessary to improve thermal prediction capability while retaining the fast thermal model required for many green data center design paradigms. This model is compared against Computational Fluid Dynamics (CFD) simulations (using OpenFOAM) and an open platform framework, namely BlueTool.