Kubeflow Operations Guide

Chapter 3. Planning a Kubeflow Installation

Planning a new Kubeflow installation is key to having a successful platform for your machine learning operations. This chapter introduces and covers the topics your team needs to consider in the process of planning your new Kubeflow cluster.

It’s important to keep in mind that Kubeflow itself runs atop of Kubernetes, and though a deep understanding of Kubernetes is not a prerequisite to following the various topics and installation and configuration steps outlined, a working knowledge of Kubernetes may be beneficial.

Additionally, though Kubeflow can be deployed on top of an existing Kubernetes deployment, the assumption in this chapter will be that a new Kubernetes deployment is being created specifically for Kubeflow. Although the narrative outlined is for a dedicated Kubeflow installation, these topics can be easily transferred if Kubeflow is being deployed to an existing Kubernetes environment.

This chapter discusses the various types of users of a Kubeflow installation, which components of Kubeflow to deploy, how storage will be allocated to users, installation on-premise versus the cloud, security requirements, and hardware considerations.

Security Planning

This section reviews the current state of enterprise security integration (authentication and authorization) for Kubernetes, and then specifically for Kubeflow and associated components. Ideally, we want to provide a holistic and fully encompassing security approach with regard ...