I really like the idea of an in-wall home control panel - the guy who converted his whole house to illustrations & animations was awesome - but when I try to imagine situations where I’d really use it, I struggle. Definitely not going to get up from the couch and walk over to a control panel when I can just use the phone app, so it’s got to be circumstances where I want to control something, while I’m in a specific place or can’t easily use the phone. Maybe walking out the door, just to see if lights or whatnot are on (if I didn’t have an automation linked to presence). Maybe in the kitchen, to fix music when my hands are too dirty for the phone, but voice control seems like a better solution there.

World is full of gadgets that seem cool, but then you get them home and it’s just easier to do things the old way, and the gadget ends up being a techy Hummel figurine.

I got an N100 motherboard with 4x GB ports and installed ubuntu. Configured 1 port for the ISP, 1 port for a guest network, and bridged the last two with a WIFI PCIe card for the trusted network. hostapd to manage the wifi, and it took a couple tries before I found a card that really handled AP mode well. dhcpd, radvd. bind 9 for DNS, integrated with dhcpd so clients can declare their own hostname. There’s a separate pihole in front of that. Took a while to discover that I had to specifically configure dhclient to the ISP to request IPv6 prefix delegation. The guest network port goes to a ddwrt router.

It also runs a mythtv backend, a couple web sites; immich, jellyfin, and paperless ngx containers. In my head, it’s more of a server with router functions glued on, where […]sense is more of a router with server functions glued on.

I was comfortable with cli administration, and the router side sort of grew organically, with a little victory each time I could unplug another appliance - first a WAP, then a switch. opn/pf-sense is probably an easier way to get into the DIY space, especially if you want to replace your whole infrastructure at one time.

My impression is that commercial loan securitization is a little different.

When you’re talking about residential mortgages, you’re talking about loans of a few hundred thousand, which is kind of too small for capital markets to bother. So the banks bundle them, like you say, and sell them in $1e8-1e10 offerings… Pension fund’s gotta park $100,000,000 somewhere, and that MBS is easier than actually managing 500 separate loans.

When you’re talking about commercial real estate - and especially things as big as data centers - it’s the opposite problem. They need such big loans that only a big bank or even consortium of big banks can get all the capital together, then the big bank turns the loan into a bond offering, so investors can buy them in slices as small as $1e4… That bond may only have one or two loans in it, so it can’t necessarily get broken into risk tranches like the MBS.

I don’t think it says anything about specific data center risks: capitalizing loans is standard operating procedure for these banks. It gets the loans off their account books, so they don’t have to hold capital to balance the risk, but they still get to collect management fees on the loans, regardless of whether they default. The big stories back in the 00s were residential real estate loans, but they do the same thing for commercial and industrial property.

All these people saying 500, 600, 700+ watts - it’s almost hard to believe. My “gaming” computer, with two monitors, peaks around 400W on graphics-heavy games, and the monitors are like 120 of that. I suppose, though, if you’re running a server-class CPU, a half dozen HDDs, and a GPU for continuous transcoding or mining, the it wouldn’t be too hard to get there.

I run a dozen containers on an N100 with a couple of HDDs, VOIP device, router & cable modem on 63W, and I’ve considered that a pretty fair deal for a home lab. Of course, that only serves the household - nothing really offered to the public.

Removed by mod

I’m happy with 90 day certs for homelab stuff: I’m not worried about anyone MITMing my network. I figure the short lived certs are more important for people who provide services with a significant external user-base, or who process sensitive transactions.

I have my certbot set to check every 12 hours and renew at 60 days. Renewing every 3 days would be 20x more load on Let’s Encrypt systems, and that feels like abuse of a free service for my use case.

I really wonder who these retailers are. Like, my grocer typically has 2 human and 12 self checkout open, and it’s no question which of those is going to be faster for someone pushing less than a full cart. They don’t even have the floor space to build twelve more full service checkouts. Reminds me of the bad old days when I’d go to the store and they’d have every human checkout line open, every one of them with 5+ people in line, and it’d take 30 minutes just to check out. If you’re a retailer that only processes a couple dozen patrons an hour, then yeah, I can see where a few self-checks wouldn’t be any benefit.

Feels like we’ve reached the enshittification phase of self-checks: big stores are now so beholden to the technology that the vendors can jack up their prices to where self checks cost the store more than a full-service checkout, but they can’t go back because there’s no staff and no space.

Tandoor definitely has a self-hosted path, including SSO integrations. It has a fair API that I’ve used to link it to nutrition data at USDA and my grocery store for prices. I suspect I could use the API to transfer a shopping list to the grocery store’s ordering system, but I don’t like the pickup or delivery systems. I’ve been pretty happy with tandoor’s import-from-web results, although the ingredient amounts often need some editing.

Also love the pi-0-2w, but I’ve started to use ESP32 for those edge sensors. You can get those for like $1-2 each on aliexpress, where real world prices on Pi zero is upwards of $20 now. Truly embedded systems like arduino were always intimidating to me, but esphome reduces ‘programming’ to a YAML for any of the many supported devices. The Pis that are working now aren’t likely to change any time soon, but I bought a handful of ESP32s for $10, and I’ll probably buy another handful before I buy another zero. Maybe convert a zero to ESP32 if I really need a whole OS for something new.

I feel like (mostly because of my own path) Pi is a great platform to get started with self hosting because of projects like pihole, homeassistant, and osmc. Projects where someone has put together simple, all-in-one solutions where you can just download an image and go. That same approach, and the Pi’s hardware limitations, make it harder to co-host multiple services, while the price point of Pi’s, at least pre-2023, made it reasonable just to run a handful of them. Only once you start wanting larger or more compute intensive stacks does the pressure to use containers/VMs on ‘real’ hardware appear. If you’ve already got pihole and ha running on Pis, there’s no pressure to move them to the big server.

If it’s the path I followed, it must be the path everyone follows :)

Calibre-web and calibre can share a library directory. You can edit the .kobo/kobo/Kobo ereader.conf file to wirelessly sync to calibre-web instead of kobo.com (while still using kobo.com to buy new ebooks)

The 1U system I had, the heatsinks have been fanless and not the extruded aluminum block one usually thinks of, but an assembled stack of much thinner, more rigid fins. It relied on airflow from fans mounted in the middle of the case, between the mobo and disks, shrouded to focus that airflow. I want to say 2x40mm fans per processor.

You should expect a 1U system to run hotter than a desktop, and definitely hotter than cooler advertisers want. CPUs are designed to operate up to 100°C or so, and the constraints that lead people to use 1U form factors generally include replacing those systems regularly, basically because of burnout.

People pay fake tickets because it’s too expensive to fight them.

Not nextcloud, but I have several containers set up to use ‘third party’ database - ie, not running inside the container or compose group, sometimes on a second machine. It definitely complicates the setup. I haven’t noticed any performance issues, but my services are low load.

I would think that the db running on the Pi with storage on the Pi could be faster than the db running on the laptop with storage on the Pi. The Pi db would presumably have higher bandwidth to the Pi storage and only have to transfer the selected data over network, where the laptop db has to do all the db sorting, filter, and processing over the network, although I suppose the db would try to keep all the data in RAM. I know there’s limitations to Pi’s internal bus bandwidth, Maybe you can compare the hdparm performance of the NAS RAID vs the NFS mount.

Goodhart’s law: When a measure becomes a target, it ceases to be a good measure.

Used to be that a college degree indicated people with time, money, and drive to put four years into study. It was a good measure of the “right” kind of person. Now it’s just another job requirement and doesn’t indicate anything.

Just what I was looking for. Recently got a kobo for experimenting, and just starting to figure out how to integrate it.

I’d still rather physical books, but sometimes there’s just too much premium for too shitty of POD trash.

These are computer monitors, not TVs. Plugging them into a Windows box triggers Windows to silently install their “driver,” which silently installs a handful of other startup apps. It’s an escalation of the smartTV-wifi path.

I like Cory Doctorow’s take: AI is good for single-use, personal code to solve an immediate problem, and terrible for long-term, production projects. I imagine there’s a bunch of neophytes out there who use AI to create their first project, find out that github exists, and thinks someone else must be having the same problem they just solved, so why not release it to the public?

Dunno how well it would work with kubernates internal networks, but my DNS is configured with different views for internal and external clients. So, when letsencrypt does a lookup, they get the external IP, but when an internal client looks up the same name, they get the internal IP. TLS is happy, because the certificate matches the name. I’m happy because it works even when the ISP is down.