A Data-Driven Approach to Monitoring Colocation Data Centers

• Published in: 2019 IEEE 5th International Conference on Big Data Intelligence and Computing (DATACOM) pp. 234 - 241
• Main Authors: Setz, Brian, Rao, Subrahmanya VRK, Lazovik, Alexander, Aiello, Marco
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.11.2019
• Subjects: Atmospheric modeling; colocation data centers; Computational modeling; data center monitoring; Data centers; Data models; data-driven modeling; Servers; Temperature measurement; Temperature sensors
• DOI: 10.1109/DataCom.2019.00043

A colocation data center rents out rack space and equipment to third parties, while providing core services such as power, cooling, and bandwidth. Energy consumption, power distribution, and cooling account for the majority of the operational costs. Reducing these operational costs is important in order to increase the profit margins of the data center. When implementing cost-saving measures, the Service Level Agreements and the high Quality of Service have to be maintained. Therefore, monitoring the effect of these measures is key. In a colocation data center however, monitoring the effect on the IT equipment owned by third parties is challenging, due to limited access to the equipment. This work addresses the question: can we monitor third party IT equipment in a colocation data center, without requiring access to the operating system? To answer this, we collect 2.5 billion data points from over 160 servers in a data center, monitoring the server state and the environmental parameters. We utilize this dataset to discover and train multiple models that allow colocation data centers to monitor third party servers without requiring access to the server's operating system. These models enable Data Center Operators to monitor the effect of cost-saving measures on the thermal state of the servers. As well as to monitor the computational load, in order to assist in the expansion planning process.