On-Prem Cloud Platform for AI Workloads

This repo contains the engineering notes of an On-Prem Cloud Platform for AI Workloads with Oracle Linux KVM, Oracle Linux Virtualization Manager (OLVM), Docker, Podman, Kubernetes (specically, Oracle Cloud Native Environment) and MLOps and DevOps tools (e.g., Kubeflow, ArgoCD) on top of the AMD EPYC 9655 CPUs and Nvidia H100 GPUs.

Overview of the Platform

Index

Virtualizing the Nvidia H100 GPU in Oracle Linux for KVM VMs

Virtualizing the Nvidia H100 GPU in Oracle Linux for Podman Containers via MIG Instances

Creating LACP-mode Network Bonds to Boost Network Performance and Reliability

Provisioning Ceph Block Storage to OLVM via NVMe/TCP

Backing up Ceph Storage to TrueNAS to Amazon S3

Enabling VNC Encryption via TLS

Getting an SSO token from oVirt Engine to interact with its REST API for automation tasks

Migrating VMs from VMware vSphere to OLVM with virt-v2v and Oracle Linux Automatoin Manager (i.e., Ansible AWX)

Migrating VMs from Sangfor Cloud Platform to OLVM with virt-v2v and Oracle Linux Automation Manager (i.e., Ansible AWX)