Deployment of Containerized Deep Learning Applications in the Cloud

Abstract

During the last years, the use of Cloud computing environment has increased as a result of the various services offered by Cloud providers (Amazon Web Services, Google Cloud, Microsoft Azure, etc.). Many companies are moving their data and applications to the Cloud in order to tackle the complex configuration effort, for having more flexibility, maintenance, and resource availability. However, it is important to mention the challenges that developers may face when using a Cloud solution such as the variation of applications requirements (in terms of computation, memory and energy consumption) over time, which makes the deployment and migration a hard process. In fact, the deployment will not depend only on the application, but it will also rely on the related services and hardware for the proper functioning of the application. In this paper, we propose a Cloud infrastructure for automatic deployment of applications using the services of Kubernetes, Docker, Ansible and Slurm. Our architecture includes a script to deploy the application depending of its requirement needs. Experiments are conducted with the analysis and the deployment of Deep Learning (DL) applications and more particularly images classification and object localization.