The most annoying thing about a lot of these is that tutorials are “minimal viable setup” sorta things. Like “now you have it setup, make sure you tune it for production”

Dude I’m already in pain from trying to serve these models and you just have to go rub salt into my eyes. “Simplify your stack with <Tech>” they said. “Share your resources effectively and easily with <Tech>” they said. “Here’s your fuckin’ ‘Hello, World’ now GRTFM and buzz off” they said.

Working close to the metal do be like that.

I work for a large enterprise and build ML model monitoring pipelines fairly frequently—this will be a more in depth but similar use case to what you’re asking.

We use Grafana (visualization) and Prometheus (timeseries db)—they’re built for this use case exactly. Tons of info out there on how to build, configure, connect to your sensors, and deploy it.

I call it N Jinx. Always have and I’ll never be convinced otherwise that it’s not.

I don’t think an SSD is the right choice here. SSDs have a limited lifespan that’s majority driven by the number of writes that happen to a certain block. Reads are cheap and near infinite though.

When you’re talking about a Lemmy instance, mail server, etc. my mind thinks this is likely to be many writes with several read-once ops. This is a better use case for a HDD.

A media server that oriented towards most consumption (reading) would be better for SSD.

I like draw.io for process diagrams.