EPFL Researchers Develop Machine Learning System to Enhance Efficiency of Data Centers
Researchers at the Ecole Polytechnique Federale de Lausanne (EPFL) have developed a machine learning system, called CloudProphet, to optimize the allocation of computing resources in data centers. With the increasing demand for data center services and the lack of transparency in how these resources are used by customers, EPFL’s innovative solution aims to improve efficiency while reducing the carbon footprint of these facilities.
Data centers play a crucial role in providing virtual machines to customers, allowing them to run applications and store data remotely. However, the behavior of these applications remains a mystery for data center staff, as customers are not obligated to disclose their processes, and staff are not allowed to observe them directly. This lack of transparency poses challenges in accurately predicting the demands of these applications and efficiently allocating computing resources.
CloudProphet overcomes this challenge by identifying application processes from an external perspective and relying solely on hardware counter information for performance prediction. The system utilizes neural networks to learn and anticipate an application’s resource requirements, improving the efficiency of resource allocation in data centers.
Data centers do have diagnostic tools for the identification and performance prediction of applications, but they can only claim an improvement rate of 18%. We are achieving results that are orders of magnitude above that, highlights Darong Huang, an ESL PhD student at EPFL.
The research team’s findings, published in IEEE Transactions on Sustainable Computing, demonstrate the potential of CloudProphet in revolutionizing resource management in modern data centers. By implementing this intelligent system, data centers can significantly reduce their carbon footprint while ensuring optimal utilization of computing resources.
EPFL’s collaborative approach involves funding PhD students and postdoctoral fellows through industry partnerships. With project managers from industry offering advice on practical constraints, the researchers strive to simulate real-world applications as closely as possible.
To further validate the effectiveness of CloudProphet, EPFL plans to implement the system in their new data center project, the CCT building. By partnering with the Distributed Electrical Systems Lab, the EcoCloud Center, and the EPFL Energy Center, they aim to assess the extent to which the system can reduce the data center’s carbon data footprint.
Once the results are obtained, EPFL will explore options such as software licensing or potentially launching a spin-off company to commercialize this groundbreaking technology.
The development of CloudProphet represents a significant step toward achieving greater efficiency and sustainability in data centers. By leveraging machine learning and neural networks, researchers at EPFL are making strides in optimizing resource management and reducing environmental impact. With the demand for data center services continuing to rise, this innovative solution holds immense potential for the future of data center operations.
