I can run my entire home lab on Proxmox LXCs without usin…

Besides their utility in self-hosted projects, Proxmox LXCs are just as viable for general server experiments

Considering that Proxmox uses the uber-powerful KVM under the hood, you can deploy virtual machines for most operating systems on your server node – and this includes everything from Windows 11 and FreeBSD flavors to Unix-based platforms and Android distros. But if you’re primarily using your Proxmox home lab to experiment with Linux environments, LXCs become a pretty viable option for your DIY projects.

I used to deploy tons of VMs for my tinkering escapades before finally hitting the max resource utilization on my Xeon server and realizing that many of these tasks can be performed on their lightweight LXC counterparts. While I wouldn’t say that I’ve ditched all my virtual machines in favor of LXCs, I’ve started relying on Linux containers a lot more than I used to in the past. Let’s start with the brass tacks: LXCs share kernel resources with the underlying Proxmox node’s kernel.

Rather than virtualizing a full-fledged OS with its own kernel, LXCs run isolated processes without taxing the server PC. While this makes them slightly less secure than completely isolated virtual machines, Linux Containers work exceedingly well for deploying FOSS utilities in a home server, especially when the system in question is an old rig, mini-PC, or even a single-board computer. To put their resource consumption metrics into perspective, the weakest node in my setup – a Lenovo G510 from 2014 with 2-core i5-4200M and 4GB of memory – can run well over 10 LXCs, with a couple more nested inside a CasaOS instance.

In contrast, the same laptop would struggle to run a GUI Debian virtual machine, and could only manage a single CLI-based DietPi VM. But once I opt for better systems, I can easily run dozens of LXCs without hitting half the CPU utilization on my server node. Of course, the type of distro and packages running in the LXC matter just as much, with Alpine-based containers being the lightest of the bunch.

Even if we left the resource consumption aside, LXC templates typically only hog a few hundred MBs in my storage pools, and I don’t need to allocate dozens of GBs to install the distribution onto a virtual drive when creating a container. Their startup times are just as rapid, and I get the same PBS-based snapshot provisions as typical VMs. When I first got into Proxmox, I was already familiar with KVM-based virtual machines, but I had no clue how Linux Containers worked.

Fortunately, LXCs turned out to be extremely easy to get into – to the point where I’d say that they’re far more accessible than their virtual machine counterparts. For starters, the Proxmox VE Helper-Scripts Repo is chock-full of LXC (and a handful of VM) templates, and these span everything from simple Linux distros and common media servers to full-fledged automation services and somewhat weird self-hosted utilities.

Or, you can go down the DIY route with a simple TurnKey LXC template. Call me a vanilla Linux user if you must, but I end up relying solely on Debian LXCs when I need a stable container for my experiments. However, there are just as many distros available in these templates, so I can switch things up with Fedora, Alpine, or even Gentoo if I so desire. Plus, a barebones LXC works pretty well for some ingenious (or rather, insane) experiments… Although LXCs are pretty handy for deploying lightweight FOSS applications inside isolated environments, there’s a lot you can do with them.

For example, I’ve got an Ollama LXC at the center of my home lab, and it’s responsible for granting the superior reasoning capabilities of my local LLMs to my self-hosted app stack. Since Proxmox also supports PCI passthrough for LXCs, this Ollama container is connected to my old Pascal card. Better yet, multiple LXCs can share the same GPU in Proxmox without requiring SR-IOV shenanigans. But it’s far from the most important virtual guest in my home lab.

That honor goes to my Tailscale subnet router that, despite requiring elevated access privileges to the host’s network devices, lets me control my entire home lab from remote networks without manually adding every LXC, VM, and server node to my tailnet. While there are a couple of caveats to running Docker environments inside LXCs, you can still configure them and get reliable performance with nested containers.

Heck, I’ve even seen folks create full-fledged gaming LXCs, though it’s something I’ve yet to build myself. Truth be told, unless I want specialized operating systems, better isolation provisions, or specific features available in the VM instances (looking at you, Home Assistant), I’d typically opt for the ultralight LXCs even on my hardcore workstation nodes.

Summary

This report covers the latest developments in android. The information presented highlights key changes and updates that are relevant to those following this topic.

---