The goal of this session is to overview, outline, and explain the technology to deploy & provision a scale-out kubernetes cluster for running & orchestrating containers & VMs on bare metal. Much of the kubernetes ecosystem is focused on cloud & shared infrastructure platforms, but there are certain customer use cases & characteristics that simply require bare-metal nodes - in fact, even cloud platforms continue to run on bare-metal servers one way or another ;)
This presentation will present our experience taking a high-density OCP rack with over 1,000 compute cores and 9TB of memory from bare hardware to running a serverless demo app in 10 minutes. We will outline the choice of OS components, network connectivity, deployment of master & shared service nodes, and parallel deployment of compute nodes, preparing everything to process application traffic.
There are many layers to the kubernetes deployment stack, including infrastructure components built on other, lower level infrastructure components - sometimes four and five and six layers deep. This provides convenient abstractions for different engineering, deployment, and service levels - but also results in a large amount of total deployed software when considered end-to-end. Deploying such clusters on racks of bare metal gear - whether OCP or otherwise - requires preparation and planning, but does not have to be a months-long endeavor. In order to support customers and SPs that require scale-out on-premise hardware solutions, we have had to evaluate and optimize various approaches to bare metal deployment and will share our learnings here.