Ah but the rub is that it would not be unreliable.
One of the things I discovered was that "reliable" in the eyes of non-engineers isn't the same thing. I have a NAS system for my important data because I know it is reliable, I know dozens of people who "have a copy on a flash stick" because they consider that reliable.
You can build "good enough" in a rack that will serve a LOT of clients. One layer of ZFS (1U + 3U) for storage, Five 1U servers, one 2U 10G switch.
In my experience that is 1 failure (that results in an outage) every 2 years at most. When I was at Google I (like everyone) had access to failure information about equipment in the "fleet." Motherboards do fail, but VERY rarely, disks are the worst offender but with double parity ZFS you have a week at a minimum to replace a failed drive. ALL the drive failures at Blekko announced themselves before they failed with SMART data.
So in addition to the colo cost you've got an engineer spending part of their time watching the rack, fractional head count cost.
I have a tiny VPS that costs less than $2 per month. I run a chat server on it. In seven years it has had no failures and one hour of planned downtime. That's better than any of the major cloud services. Simple stuff is surprisingly reliable.
I’m guessing that if you don’t join big federated chatrooms on your Matrix server, it runs absolutely fine on minimal resources.
One of the smarter things the bluesky team looks to be doing is to split the world into heavy-lifting versus personal servers, so folks don’t end up with their personal servers suddenly sucking up lots of resources just because a user views #wtf or whatever. We should look into this sort of
tiering too for Matrix, to eliminate the “Matrix is resource heavy” snark.
I don't federate. If I had more control over how much data other servers could dump on my server, other than restricting rooms with X number of users, I would. Synapse and Element together are heavier than any other chat servers/clients I've ran, with only one exception.
It is not a big problem, because my number of active users is fairly small (under 50), but I could see myself having to use a different server software, or possibly a different protocol all together, if it were to open up and grow.
And if that vpn had been hosted in OVH's Strasbourg datacenter, in 2021 you would have had a huge downtime and possibly total loss of data without a backup in another datacenter.
It is like driving under influence. Many people get away with it most if not all their life. Other die in stupid car crash alongside their pregnant wife mere days after their mariage.
