logo

Your Own Little HA Cloud at the Edge (or at Home)

Conference:  CloudOpen 2022

2022-06-22

Authors:   Stephane Graber


Summary

Building your own infrastructure can be cost-effective and less complex than using cloud services, with open source technologies like Ceph, Lex, and Oven.
  • Cloud services can be expensive and complex, with potential vendor lock-in.
  • Building your own infrastructure can be cost-effective and less complex, with open source technologies like Ceph, Lex, and Oven.
  • These technologies provide redundant distributed storage, network compute, and frequent releases.
  • They can run on cheap hardware, such as Raspberry Pi.
  • The speaker has tested these technologies in a data center and found them to be stable and reproducible.
The speaker recalls the days when people would boast about their server's uptime, but this often meant neglecting security updates and being afraid to reboot the machine. Building your own infrastructure can avoid these issues and provide more control over the system.

Abstract

Who wouldn't want to own their own little cloud, running only their own instances on their own hardware, yet still benefiting from a highly available environment including distributed storage and networking. This is what we'll be exploring in this talk, combining LXD as a self-hosted mini-cloud with CEPH for storage and OVN for network. That combination allows for a fully redundant deployment on as little as 3 machines and can run hundreds or thousands of containers or virtual machines, be managed from anywhere and run just about any workload you can think of (GPU and all included). Whether you want something like this as a homelab running on some Raspberry Pis or running it in a co-location facility with a few small servers or deploying it at the edge, you always get the same thing, something which feels like a cloud but that you fully own and can easily reshape to your liking. This talk will start by going over what LXD, CEPH and OVN are, how they can work together to provide a highly available deployment on as little as 3 machines and then will go over how to actually deploy this in just a few minutes on low cost hardware.

Materials: