Telemetry Analytics to Secure Container-Based Cloud Computing

Containerization, based on operating system virtualization, provides high scalability and mobility and is currently adopted by many cloud service providers to lower infrastructure and maintenance costs. However, container-based cloud computing systems reveal broad and unknown attack surfaces because container servers share vulnerable kernel features between independent, highly scalable services which generate complex network activities. To protect container-based cloud computing systems, we will develop telemetry analytics that use deep learning models with pretrained networks. The telemetry data contains temporal order as well as spatial dependency, benefiting both convolutional neural networks and recurrent neural networks. These combined deep learning models will be trained not only using data collected in each layer of the cloud from infrastructure to applications but also using pretrained networks previously trained on traditional cloud data to enhance analytics generalization. The final goal of this project is for the proposed deep learning analytics to enhance detection of new and variant attacks specific to container-based cloud computing.