Temperature variation impact on estimating costs and most critical components in a cloud data centre

Abstract

Cooling plays a very important role in data centre availability by mitigating the overheating of Information Technology (IT) equipment. While many existing works evaluated the performance of the cooling sub-systems in data centres, only a few studies have considered the important relationship between cooling and IT sub-systems. This work provides efficient models (using Stochastic Petri Nets (SPNs)) to represent a cooling sub-system and analyse the impact of its failures in terms of service downtime and financial cost. We provide an efficient model, diminishing the state explosion problem when modelling DC availability. The study also identifies the components that are the most critical with respect to service availability through the use of sensitivity analysis. Results show that the adoption of a redundant architecture reduces the annual costs related to downtime by as much as 70%. The chiller is observed as the main component that affects service availability and operation costs.