New Storagenode build - Intel N100?

Skip the power-protection workarounds. I’d just accept that I may occasionally need to fsck things if I got unlucky with a stray write during an outage.

Set the BIOS to always-boot-when-power-restored, then slap in a cheap PCIe KVM so you have remote console and power control (or if your one PCIe slot will be filled, use the external Cube). 99% of the time the system will boot itself fine when the power is restored. But when there’s a problem you know the KVM will always be there (with alternative boot ISOs if you need them) to get you in to fix a filesystem.

This does mean you’ll need 2 Ethernet connections instead of one, but that’s probably no big deal.

I tested lot of batteries, some are very expensive and shitty, dont deliver what is written on them. I use Kung long WP1236W 12Volt 36W it is 9AH battery in case of usual 7ah
and it is not very expencive. Also always pay atension when you buy, when was battery made, it usualy have marks on battery if it made more than 6-8 months ago dont take it.

1 Like

I’ve had a good experience with PowerSonic batteries. They tend to last way longer than expected 2 years (3-4), but I still replace them in the UPS attached to important equipment (my storage servers) every two years and move them to an ups attached to less important ones (modems/routers/switches)

1 Like

Yes know them also they are also OK. usually it my second choise.
I know CBS are very good batteries, but very expensive.

The LiPo sounds interesting, I believe they won’t suddenly die on you like acide ones. Do you know of some models?
APC smart-ups 750va SMT750I after LESS than 2 years (I remember I checked when the battery was changed, with OEN one at vendor support facility), when the power from the grid shut down, my server shut down abruptely, like there was no UPS.
The power draw is less than 100W. When the battery was new, it could keep the system on like 15min or more.
Now I use only Cyberpower CP1500EPFCLCD and 1300… The oldest ones are from 2021 and they keep the nodes on more than 10min, even the server works enaugh time to shut it down safely.

Whoa that’s far from cheap. That’s pretty much half the cost of the whole motherboard (with IPMi) and a processor

Basically, it’s a consumer product half-arsing the problem that is already solved in enterprise chipsets.

Plus, buying used, you are keeping hardware in service for longer. It’s a good thing.

Small switches are $5 on eBay too :).

I don’t understand the use of a KVM that I just discovered with your link.
I get that it’s a mini PC that you remotely access, and you can WoL the main PC, but other than that?
The system is configured to boot up when power is restored, I have remote access in the local network through Wireguard that runs on my beautiful Asus router :blush:, and I can WoL the system if I choose this way, or let it boot by itself, and SSH with PuTTy to it.
What KVM can do for me extra?
If a power wave knocks off the router, than no KVM access also; if a power wave knocks off the mobo, than no KVM access also, and if a power wave can frie other components, than for sure can frie the KVM’s power source.
What do I miss?
A chineese KVM is just another backdoor into your LAN for the asians.

Whoa that’s far from cheap. That’s pretty much half the cost of the whole motherboard (with IPMi) and a processor

It’s half the price of a RAM-less, CPU-less motherboard that came out 10 years ago. I love old enterprise gear as much as the next person, but LGA1151 very much has it’s limitations in todays environment.

1 Like

Sure. But for storage server? It’s still more than enough. My main storage server is x9 motherboard with E3-1230 v2 (that’s sandy bridge! Not even PCIE3!) and cpu utilization rarely exceeds 10%. It’s clocked down, under-voted, and will work forever. I’m yet to encounter its limitations using storage at 2.5GBps. It’s more than enough.

CPU is included. Ram isn’t. You can ask them for ram they’ll throw it in at rock bottom price.

1 Like

For a storage server? Yeah sure. I also use a v2 xeon chip - specifically E3-1220L v2 in my case. It’s a pure storage server, and while IO-wait is rather high, CPU utilization is at around 20-30% at all times. I agree - It will work forever, and I love it. I will swap it out, when it no longer works. That being said, I don’t think I’d get the same board, if I had to buy today.

I bet you’ve had your board for some time as well, and is probably in the same position. We’ve been in a similar discussion before - the “right” way to do things is to have a pure storage box (pure in function, not the brand), a pure application box and pure networking on top of that. We both follow this approach, but I think the vast majority of users want’s a fractal design case that the wife can approve of to sit in the corner of the room and do everything

It gives you access to early boot options, and power/media control.

  • Like if your boot partition needs a fsck after a power outage: or it tries to check a HDD and fails, and it drops into single-user mode. Can’t SSH in when it hasn’t even booted to networking yet.
  • Or if you have halt-on-boot set for SMART errors… and after a restart it sits on the BIOS screen saying “SMART error detected on device X: Press [F1] to continue” or something.
  • Or you want to reinstall from Ubuntu to AlmaLinux. Or just need to temporarily boot into a live recovery distro
  • Or you’re trying a new kernel update… that doesn’t work and the reboot fails. You need grub access to go back
  • Or just the system locks for some reason. Need to hold the power button down for 5 seconds (or hit reset) to revive it.

Basically KVMs can make it like you’re sitting in front of the system. You can see the BIOS. You can add/remove boot media. You can hit the power and reset buttons, and see the power LED (and sometimes HDD LED).

But yes… if the system is alive-enough, usually you just use SSH. And like AR mentioned many enterprise-class systems come with built-in remote management. It lets you take almost any consumer system… and for a few $$$ add that remote management.

3 Likes

I understand. Yes, it seems useful to have one, but there is a big risk that these devices are backdoored. I don’t know enough to be able to spot a shaddy connection. Maybe if there are KVMs from known reputable manufacturers, not based in China…

They’ve open-sourced their respective baby-Linux distros (and at least NanoKVM is porting the popular PiKVM software too - not sure about JetKVM). But yes… like absolutely any other device… there could be hardware-level exploits :wink: .

What’s wrong with being based in China?

Exactly. Or start with enterprise one to begin with, where the BMC (board management controller) is properly built in, and time tested, and definitley works, so you don’t have to hack ugly half-working solution as an afterthought, dealing with rats nest of cables, and avoid using consumer parts altogether.

If you are concerned about backdoors — configure firewall. Only you shall be communicating with the BMC. Trying to spot malicious connection is a bad approach. It’s much easier to block everything. Then you don’t need to monitor anything.

Anecdotally, mine cannot communicate with anything but a small arm computer in the lan, that runs cloudlared. Then I can access my machine literally at https://truenas-ipmi.arrogantrabbit.com from anywhere. (You can’t reach it unless you steal my github account, but I get to the UI directly, as I’m already logged in). Few months ago when a boot drive died (I use shitty thumb drives) I have reinstalled TrueNAS during lunch while at work. Highly recommend.

Then I have had a best node ever. Pi3 in the sofa. Quiet, and do not break a design :smiley:

like… in the cushions?

Yeap. It’s not because it’s not a design issue, it’s because this sofa covers the power and Ethernet ports, and there’s a big empty space inside. There should have been bedding and pillows there, but there weren’t that many.

Well, and at the same time, it solved the problem with the design (well, I couldn’t make this little thing look nice, despite the button backlight… they didn’t approve…)

3 Likes

So if some higher authority decides to change the couch, you will hear in the evening:
“So dear, I bought a new couch; but this one dosen’t plug in. I believe it’s wireless or something”.
“And… what did you do with the old one?”
“I unplugded it and took it outside; it’s raining, but it’s old anyway. We don’t need it anymore.”
…“and the dog chew the cables”. :smiling_face:

But why? The small one is sufficient. And the big one looks like now is in wrong orientation…

Remember, just 50C is unbearably hot to the touch but perfectly fine for electronics. The solution is not to touch, not to install massive heat sinks.

“sure You wanna know?” (spider-man.jpg)

Just to be sure, because the machine case is not a typical server one, and the power of fans are not server ones.
Also as Dr. Frankenstein once famously said: “because i can”
Also the orientation is correctly oriented to a custom side fans.

To overload Your curiosity i can say that ind the end, it didn’t even matters, because controller was not the problem at all (vs some non-server class one), as i later found out, nor the VM’s. The software was, as always, thanks to M$.

I found only the newest OS from Windows family are able to utilize modern HDDs to the nominal advertised speeds and read times. Disk produced in 2017 needed to wait for a OS made in 2021 to unlock all its possibilities, nice, but that what it is, and linux just wasn’t the option.

1 Like