You don’t mention your throughput requirements. How fast is your internet connection? Will you be a VPN server and/or VPN client? A reverse proxy? All that adds overhead but really not that much compared to other services. It just changes your requirements as to how many years obsolete your hardware can be.

Generally, whatever desktops or laptops businesses are throwing out in the trash will be more than enough.

If you have a managed network switch or one that can do VLANs, a router on a stick will work fine, especially if your Internet connection isn’t more than half the speed of your server’s network card. A repurposed laptop is perfect for this, because it has a built-in UPS and console!

I’ve got a 13-year-old server handling my 300Mbps internet connection, Wireguard, reverse proxy, and other stuff. It used to handle a backup internet connection, too. It’s regrettably on pfSense and I’m trying to migrate it to opnSense but my setup isn’t exactly by the book. I put stuff that’s supposed to go out over VPN on a separate VLAN and I give them a separate OpnSense router running in a VM so there’s much less chance of leakage.

One thing I did learn the hard way is that a lot of consumer “smart” devices wrongly assume to be on the same broadcast domain as any servers, clients, or peers they talk to, so even with avahi handling relaying between VLANs, they won’t work. It’s annoying having to move your dishwasher off the IoT VLAN just to make it work.

I dunno, mailcow dockerized seems to work ok for me. That being said, e-mail is so 20th century.

Depends on the currency and features. If you’re looking for something outside of the 14 eyes that allows port forwarding, your options are extremely limited.

Any momentum on this front gets me excited, even if it doesn’t personally apply.

Since it’s cost-effective to combine gaming requirements with AI server requirements, I have my multi-modal language model stuff running on my (admittedly seldom-used) Windows gaming desktop. That means running most GPU-related tasks (aside from encoding/decoding/simple object recognition, which uses a separate server containing an Arc A380… purchased before A310’s were available) in docker running under Windows Subsystem for Linux 2 (WSL2). Running stuff as background services just makes one assume that it should be a logical step to just make it multi-user. Easier said than done, I guess, just like multi-user stable diffusion.

Getting Games on Whales running under WSL2 has taken me down the familiar but unwelcome rabbit hole of recompiling Linux Kernel modules, which I’ve experienced is more straightforward on bare metal than WSL2.

The more attention and excitement about this topic, the better.

No more listing dozens of sensors when I ask what the temperature is outside, perhaps?

Funny, every time someone mentions pi-hole, I have to look up why I don’t use it, and I wonder if others do the same.

My combination of pfSense and its pfBlockerNG package does pretty much same thing and more, and once I migrate to opnSense, I have high expectations I should be able to do something similar.