Hacker News Reader

Comment by manmal | Hacker News Reader

For anybody reading this - I think it’s a great config, but would be careful around pico PSUs in case you want to run a bunch of good old spinning disks. HDDs have a sharp power peak when spinning up, and if you have a couple of them in a RAID, they might do so synchronously, potentially exceeding the envelope.

View on HN

You're right. 6 spinning disks and a discrete GPU won't work with 120W. I would recommend to get the Corsair RM550x 2021 (or similar) when using a NAS with more than 2 disks or a discrete GPU.

I personally don't need more than 10TB space, so 2 10TB Seagate EXOS in ZFS mirror would just work fine with 120W as long as you don't run Prime95 the whole time. You might go up to Pico PSU 150 then.

Even if you have no issues with spinup, insufficient power can result in silent data corruption, which can be quite frustrating to diagnose (even if ZFS will help you see it). The difference between a picoPSU and a decent "real" PSU in terms of power consumption at the wall is maybe a couple of watts in my experience - not worth the risk for most people.

Is there a source for this?

Not sure. I experienced this personally using a picoPSU with spinning drives on one machine - no issues with spinup but silent data corruption on writes (which then got picked up by ZFS). Switching to a traditional PSU cured the problem.

Ok, I have running multiple machines with PicoPSU and never experienced this. In my opinion it is a pretty reliable consumer device.

However, thank you for pointing that out. I did not know that.

Can't say anything about picoPSU, except writing anything on RAID involves all the drives to seek and write and without a a solid power reserve you can receive a momentary dip of voltage which can lead for a malformed write, but back in mid-late 2000s I've seen enough HDDs failing and corrupting data which miracluosly behaved when moved to another machine or when you replaced PSU (with a decent one, not another $15 chinesium).

A lot of bad reputation of WD from that time is actually from the shitty PSUs, not the drives themselves.

Well, that's really interesting. If I ever experience these issues somewhere, I know where to start...

to go deeper, depending on a file system, some FS won't let HDDs go to sleep, so they always consume power for max RPM.

> some FS won't let HDDs go to sleep

Which ones?

btrfs and zfs at least, probably most frequently used FSs for RAID

You can do it with ZFS at least. I have a ZFS setup which idles drives. If I don't access the datasets on the drives, they don't spin up. You do have to be careful with things like auto snapshotting tools though - that will typically wake drives up. I had to script mine to only run on relevant datasets when drives are spun up and also had to edit the tool (sanoid) to prevent it from enumerating the list of snapshots on the idle datasets not being snapshotted.

I'm running a ZFS NAS with mirrored 2 HDD vdevs. I've set the HDDs to spindown after 20mins using hdparm -S and this works fine under zfs. The drives (Toshiba N300s) take ~15s to spin up when something accesses them, and ime zfs has always handled this gracefully.

This vdev only contains data; it's not running the OS / root.

In times long gone there's been a setting.

Staggered spin-up it was called.

Turned on one disk after another.

So the PSU didn't bother, or stalled.

Done.

Reply on news.ycombinator.com