Several HDD under one sharefolder.

Mergerfs can handle that. You just need to specify "Existing Path" policies. And make sure the path only exists on the right HDD.

If mergerfs use an EP policy, then files end up on HDDs with that existing path. If only one HDD has that path, then all files end up on that HDD.

Just curious. Why do you want to do it like this?

One filesystem split up between HDDs. It seems you could use one shared folder per HDD instead?

Using the HDDs as separate filesystems might be wasteful and require manual shuffling of files as drives fill.

An option could be two mergerfs drive pools without EP policy. Mergerfs will handle the shuffling of data and consolidate free storage. But you lose control over what drive the files end up on.

Perhaps just Most Free Space. MFS. Then the files will be evenly spread out on all drives which gives more even drive usage and perhaps even performance benefits. One mergerfs drive pool for backups of the other. The drawback is that drive spindown is less likely since there will be loss locality of files. Associated files in the same folder are likely to end up on different drives.

Another option is Least Free Space. Then the drives are filled, one after another. Great to consolidate storage and easy to add a new drive when needed. More locality for associated files.

A combination might be best. MFS for shared live data and LFS for backups.

One reason why you might want control over what drive files end up on could be that you use some form of tiered storage. So that you have cold "archive" data that never/rarely change and thus does not have updated backups. And hot data that is often added to or updated that is backed up often. You only need to update cold backups when you move data from the hot to the cold tier. I use a tiered system for parts of my storage.

You may want to investigate snapraid as well.

Nice build.

I have ordered parts for a DIY 19" 18U rack. I intend to move most of my boxes there. Odroid HC2s, RPi4, switch, PC and a server built from salvaged old PC parts.

Inspired by: https://tombuildsstuff.blogspo…erver-rack-plans.html?m=1

I bought two 4U server enclosures to put my existing desktop in, as well as a OMV server built from various parts. Plenty of disk space.

https://unykach.com/en/servers/rack-case-4u19-uk-4129-51912/

Yes, with mergerfs you only lose the data on the defective drive. And only if it wasn't backed up.

I agree that mergerfs may complicate restores. But using rsync scripts it is possible, and even easy, to selectively restore missing files here and there from a backup snapshot.

You remove the defective drive from the pool.

Then, provided there is room in the pool, you essentially "invert" existing rsync snapshot backup scripts. So instead of updating the snapshot backup from the pool you update the pool from the latest snapshot.

(I believe snapraid provides a similar functionality?)

And if the defective drive could be fixed, you can retire it to a cold backup pool. And later add a new fresh drive to the pool to increase the free storage.

I backup everything at least once. More typically two or three times. Sometimes more.

By distinguishing between cold and hot data I can greatly reduce what I need to backup often. In addition I use rsync snapshots for my "hot" data, meaning that only new or changed files need to be updated in new backup snapshot. And I typically keep 7 daily, 4 weekly and 3-12 monthly snapshots. And the rsync snapshots run automatically every night. At least.

I will mostly reuse my existing old computer hardware. At least initially. My goal is to build a really quiet server rack with a lot of storage and a lot of internal bandwidth for fast backups and restores. And big fans spinning slowly and quietly. I don't really need a lot of performance. I am mostly happy with the performance of my HC2s and my RPi4, but I want a AMD64 server in the network as well. Not everything will run on ARM.

The Odroid HC2 is a 8 core ARM32 SBC with a SATA III port and GbE. Low power and no fan, but a big aluminium cooler that is also used to mount the HDD and allow stacking of several HC2s.

You can check out my old thread here: My new NAS: Odroid HC2 and Seagate Ironwolf 12TB.

The HC2 is a bit dated now, given newer and more powerful ARM64 SBCs.

It is hard to compare directly but I would say it equal to, or outperforms the RPI4 in raw CPU power. The HC2 has GbE and a SATA connector for a single 3.5" HDD, using a fast USB SATA interface. But is headless and only has a USB 2 port. So it is very specialized for use as a single HDD NAS. Indeed, HC stands for Home Cloud...

This makes the HC2 into a very efficient and specialized way to make the storage of a HDD available on a network. It can easily saturate a GbE connection. Also it can run dockers like Plex or Emby fine, but not transcode 4K in real-time.

I have several, each with a 12-16TB HDD. And each is connected to all the others using a GbE switch, autofs and NFS. So all the HC2s can simultaneously access all files on all HDDs for media streaming, backups or restores. As can clients like PCs, laptops, phones and tablets.

Zitat von yakelkun

why such a heatsink, is it intended to cover a 3.5 HDD ?

Yep. If you only want to use a 2.5" hard drive, get an HC1.

That screw with vibration decoupling looks very interesting. Do you have a source you are willing to share?

Zitat von Adoby

That screw with vibration decoupling looks very interesting.

I've had a couple of Lian Li cases come with something similar but not sure where you could get them otherwise. Seem like they wouldn't help the odroid-hc* since the drive sits on the heatsink.

I would use the screw for normal PCs, not for a HC2. I cut open some soft silicone tubing and put that under each stack of HC2s.