Efficient sensor placement in HPC facility for hotspot detection and server node exhaust air temperature prediction

Power utilized by Data centres and High Performance Computing (HPC) facilities are growing at a very high rate annually, the major consumer being the cooling system. Efficient thermal monitoring is required in order to avoid over/under cooling of the facility. By efficiently placing sensors inside the facility, hotspots and over cooled areas can be identified and corrective measures can be initiated. In this paper modeling of an HPC facility was done and Computational Fluid Dynamics (CFD) analysis was performed for identifying the areas, were probability of hotspot generation is high. Wireless Sensor Network (WSN) network was used to collect temperature data from that area. We are proposing a new method for efficient placement of sensors, and for node exhaust air temperature prediction. Placement of sensors was determined using correlation analysis of the collected data. After the placement of sensors, server node real-time exhaust air temperature for the monitored nodes was predicted for over cooled/hot spot detection, using regression analysis.