Not Every Idea Was a Winner
It’s okay to quit what you started, kids
I’ve quit more projects than you’ve even started. I don’t mean that as a flex, that’s just how homelabbing goes. I’ve had a lot of big ideas especially since I started buying SBCs, but not everything I started worked out as intended. When I start a project, I do some very cursory research to make sure the idea is at least technically feasible. I’m talking broad strokes - gloss over the docs, check Reddit for actual user experiences, watch some YouTube videos, etc. Then I just dive right in because that’s how I learn best. Sometimes the decision to abandon the thing comes earlier in the deployment, sometimes later. Or sometimes that decision comes much later, even after everything has been running fine for a while. Ultimately the process of vetting out something is just a matter of testing and time. Here’s why I abandoned some of the projects that I did.
It did not work as intended or advertised
Exo - I heard about this awesome thing called Exo, which is clustered AI software that allows you to combine the CPU and RAM across multiple devices to host an LLM solution. According to the documentation, you just install it and run it on all your devices and it works magically. The devices recognize each other over the network and conspire to do your AI bidding. Nope. I ran into errors, spent a bit of time troubleshooting because I really wanted it to work but ultimately got nowhere.
Photoprism - ah, the idea that sparked a homelab revolution. If you read the tale of my Pi obsession you’ll recall that my first Raspberry Pi purchase was inspired by my idea to perform photo sorting magic on my picture library using Photoprism. Eventually I got around to it and the disappointment was laughable at best. It touts that it’s capable of “automatic classification of pictures based on their content.” - tags, effectively. Wrong as fuck. I gave up almost immediately after it failed to tag any test pictures with cats in them. That’s, like, most of my pictures lol. I’m still sort of shopping around for a solution actually.
I did not understand how it was supposed to work
Hailo AI - sometimes I miss some important information during the initial ‘broad strokes’ which, had I seen said information sooner, would have caused me to find another solution or give the project up. The Hailo AI kit for the Raspberry Pi is one such case. I was so enamored by the shiny, new thing that I didn’t really do much research before purchasing. I blindly came to the conclusion that in all its AI omnipotence the AI kit could automatically do all the things I dreamed up for it, like work with Photoprism to do photo analysis and enhance the capabilities of local LLMs.
Okay fine, I didn’t do ANY research because if I had I would’ve learned that the Hailo 8L doesn’t do either of those things. I really wanted it but now it’s just sitting in a drawer in my lab - big facepalm.
Hardware limitations
I realize there are more hardware-limited ideas here than these two, but the difference is in my expectations - I would argue both of these projects should’ve worked better than they did:
Digital photo display - When you search the web for Raspberry Pi Zero projects, a digital photo display is on basically every list (including my list). Want to know what none of them tell you (including my list)? If your pics are over 1 MB in size, the Pi Zero flat out cannot keep up with rendering them during any reasonable digital picture frame transition period.
Wifi extender - I was between projects on one of the Pis and one of the kids was complaining of the wifi signal being weak in their bedroom (on the other end of the house and on a different floor than the router). I thought I’d try building a wifi extender since that’s a thing that people do. I knew there would be a performance hit because of the Raspberry Pi hardware but I was not prepared for 1996 56k speeds - no joke. Granted, I was testing using an old 54 mbps USB wifi adapter because the new wifi 6 adapter I purchased was not compatible with the Pi - more fine print I missed out on SMDH. Someday I want to build a travel router, which uses the same hardware setup but obviously I need to find viable hardware first.
It’s too complicated to deploy or maintain
If there’s a million things to configure I usually give up and look for something easier because I’m lazy and I have no shame. Here are some of the things I deemed too difficult to use:
Ansible - oof infrastructure as code. I can write a YAML file, I do it all the time with Docker Compose. Ansible takes that to the Nth level by combining multiple files and creating dependencies and playbooks, and now you’ve just really lost me. Overall I would consider my homelab a collection of pets rather than cattle. Besides system updates and a couple common apps among them, they all have different purposes, software, and hardware. Because of this it’s hard for me to justify sitting down and learning Ansible in earnest, purely for my homelab.
Frigate - this is one powerful tool… that took me months just to get running, only to discover that the Pi 5 just cannot keep up with more than one camera feed. I temporarily had a use for the Hailo 8L card in that I had a working configuration with Frigate. I considered that a big victory in itself because the developer had just released a beta version of Frigate compatible with Hailo at the time I was testing. But that’s as far as it went - I got it working. Frigate is extremely powerful but unless you want ultra-granularity into your security camera motion detection, and have a powerful system for supporting more than one camera, go with something else.
Dietpi - but not for the reason you might think. Diet Pi is a super-minimal OS built from Debian and optimized for SBCs. Man, they took out everything. Granted, a fresh install runs consuming only about 165 MB of memory in my testing, there’s effectively nothing left in the OS but necessary OS components and networking. At every corner, everything I wanted to do was missing some app for it. SSH, SMB, SCP, Git, among others. I really wanted to love it but it just got annoying. Though it could be a sign that I’m doing too much with my SBCs… lol jk jk.
The idea is over-architected
openmediavault - Sometimes less is more. But other times you don’t know less is more until you’ve tried more for a while. The latter is the case for my experience with OMV. This is a feature-rich self-hosted NAS application with loads of functionality that easily rivals expensive off the shelf, big name NAS solutions. OMV can manage disk operations, create network shares, and run software like Plex or Pi-hole in containers, and even perform system operations like apply updates and manage users. Mind you, I ran OMV for months and only ran into an issue once - I borked it after an update and I had to rebuild it but that was likely more to do with my lack of Linux knowledge than a problem with the product itself. Indeed setting up OMV as a home NAS was the first real useful thing I did with a Raspberry Pi.
Why did I stop using it? Because at its core OMV is just a handy UI that does things you can do (and may already be doing) from the command line. It was great when I was first getting started with Linux because I didn’t know what the hell I was doing but by the time I deployed a personal cloud, OMV would’ve been left running a couple containers. So I took it down and now the server just runs my critical services in standalone containers, OMV would’ve just been extra.
If you can glean anything from this, it should be that while hosting your own things can be fun and challenging and rewarding, you gotta know when to give up on the dream sometimes - to save your time, your energy, your money, your mental well-being in some cases. You heard it here first: it’s cool to quit.