Get rid of the RPi. They are not reliable (I know plenty will disagree with me here). My HA has been super reliable ever since I ditched the RPi and got a NUC (and I have heaps of ZWave too).
I would second this. I had my HA running on an RPi 4 and had mariadb running on my NAS, but ended up ditching it for an old NUC (5th gen) that I had lying around. Getting HA OS on it wasn’t the easiest (required booting Debian live and flashing the SSD) but in the end it is rock solid.
Watch this: 6 Hardware Recommendations for Home Assistant - YouTube
I also started on RPi4, and as my system got bigger and bigger, I just moved it to an Quadcore Intel Celeron based mini PC with 4GB RAM.
This I consider one of the best moves, I could find.
The difference in reaction speed was huge! Rock solid now with around 30 Z-wave, 30 WiFi and 30 MiLight devices.
I can not say it hard enough - Just try it…
I keep hearing things like this, and I I’m sure it’s true for some people, but it’s very different from my experience.
As I said above, I’ve been unable to tax my RPi B3+ in any way with HA. It runs happily using almost no CPU and very little memory. Every response is instantaneous. I want to upgrade to a Pi 4 but can’t come close to justifying that, never mind the cost of a NUC or the bother of having to build and maintain an OS for HA on some old hardware.
The great thing about HA is that you can do so many different things with it. Obviously some are going to require more horsepower on the host hardware than others. Use what works best for you. I just don’t want to see anyone driven away because of blanket statements about what you shouldn’t do.
If you keep hearing about something, then at least a bit of it could be true…
Exactly. Some users have reported improvements in speed switching to faster hardware. I totally accept that, for them, it’s a great improvement.
For me, it would do no good. It won’t turn my lights on any faster than instantaneously. Much as I love tinkering with hardware and software, I can do plenty of that without upgrading my HA host.
I’m not saying everyone “should” do what I do. Nor am I saying HA can never run better on faster hardware. But it is most certainly not accurate to suggest HA doesn’t run well on a RPi.
Same. I don’t get it either. I’m running a Pi3B too and I have a rather large installation. My CPU is at 3% on average, there is almost no IO on the SD card (database on NAS). I don’t know what a faster system could possibly improve here. Reaction speed is already instant. Make it, uhh, more instant ? I really don’t understand how people can max out a Pi4 (which is considerably faster than a Pi3) unless they’re running heavy CPU intensive stuff like ML or transcoding camera feeds.
The difference is that I run a bare metal HA install without Docker or supervisor and with only minimal integrations (but the MQTT broker is very heavily used, it’s hosted on the same Pi). From various posts on these forums it seems that a full HAOS install is a lot more taxing on resources. I get the feeling that all this Dockeritis and supervisor stuff adds a lot more overhead than most people think.
But if changing the hardware improves peoples user experience, then all the better !
That could be quiet the difference with an “out of the box” install, where everything happens on the same machine. Could…
But, hey, for me the idea is the same. I don’t say, Pi is bad (use more of them for different things). Why the move to an Intel platform did made my life easier I will never find out. It just happened…
As long, as someone’s install works great on a Pi, it’s just perfect.
There are so many way’s to get to Rome
If you leave the db on the SD card, then you’re in for a world of pain. Not only in terms of speed (SD card i/o is very slow, especially on the Pi3) but also in terms of reliability. But offloading it to somewhere else or using an SSD should solve that. No need for an Intel Core i9 with 16 cores to toggle your kitchen lights
But of course, if you feel your experience improved with better hardware, then great ! And there’s certainly things you can now do that a Pi can’t (like the aforementioned AI).
I just wanted to say that you can have a pretty complex setup that runs perfectly fine on a Pi. You might have to approach installation differently in order to avoid unnecessary overhead. I never used HAOS or Docker based HA installs, so I don’t know where the difference comes from exactly. I suspect it’s an accumulation of lots of small things that stack onto each other and slowly but surely bog the system down.
That was my initial concern. I didn’t want to fuss with getting my RPi to boot from an SSD, but having the db on another system did cause holes in my data on occasion during maintenance. I HAD an old NUC available so the cost was minimal to switch, though I do miss running HA off the POE Hat on my RPi.
The most obvious difference with not running on the Pi, like the most noticeable, and measurable difference, for even the most minimal install with no addons installed. Is that a restart of Core, is massively faster.
Yep I moved my HA onto a vm and its been rock solid since installed
yeah, i wait like 1 week before updating, so i dont get into panic mode with all major issues! i am running 12 apartments full of devices! and off course! SCREW Z-WAVE! i had some major issues with that! just went all wifi and knx for now
@DaftHonk, makes perfect sense, especially if you already had a NUC around.
Yeah I mean it certainly is. But then again, how often do you restart HA ? On my Pi3B:
Not blazing fast, but for the maybe 4 or 5 times I restart HA per year, I can manage 17 seconds
Less often now that a lot of things can be reloaded without a full restart, but when I was using it on a Pi, it was before all the improvements had been made. So yes, I was restarting multiple times daily - and each restart was taking about 5 minutes before the system was fully up and running again. It was closer to 30-40 seconds when I moved to a virtual machine.
Holy… Woah, yeah I can understand how this would be really annoying after a few restarts. How complex is your setup that it took you 5 minutes to start
Strongly advised for USB3 platforms
Not very, the restarts were to add MQTT sensors, and template sensors mostly. Even now, it’s still not that complex.
MQTT (broker is on an remote Pi), Zigbee(via MQTT) (but via a remote Pi), ZWaveJS (but via a remote Pi), Onvif Cameras, Doods (remote Pi). So Home Assistant itself is largely decoupled from the actual processing of anything other than state updates.
Sounds similar to my setup (lots of MQTT and template sensors, zwavejs, cameras).
I guess this is getting a little offtopic here, but there is something I really don’t understand. A lot of people here on the forums mentioned how their startup times were very long on a Pi and their UX generally slow and unpleasant. They experienced a significant performance increase when switching to better hardware, as you clearly did too. I always assumed that these people had ridiculously complex setups, because I was never seeing problems like this on my installation. But there has to be another reason somewhere. Between my 17 seconds and your 5 minutes startup time… Could it really be the install method ? But what makes yours so slow then ? Unless your setup is about 18 times more complex than mine… Anyway, just curious and probably not the right place to discuss the details of this.
I think there must be a fundamental difference somewhere that would explain why some people have no issues running their HA on a Pi, while others struggle a lot.
I was working on the theory that it was the ridiculously huge database being read from the SD Card that was the impacting factor. Startup times were certainly better when I moved the database away from the SD Card to an external MariaDB server.