What do all of you think about #unraid (aka. nonraid)?

Would you recommend using it? Has it some not that obvious caveats and pitfalls one should be aware off?

Edit: Here the Reddit thread link btw, almost forgot: https://www.reddit.com/r/DataHoarder/comments/1ru42d2/how_to_best_use_unevenly_sized_hdds/

#storage #homelab #selfhosting

Oh and follow up question, anyone using "MergerFS with SnapRAID" instead?

And one person was suggesting #BTRFS, not sure how good of an option using it on a bunch of unequally sized drives is considering that RAID56 is still considered unstable.

However if I wanted to store multiple full copies that probably could work though.

@agowa338 from a long line of failed sata and sas controllers and failed motherboards, failed drives, some running unraid - technically RAID4, I recommend only RAiD10. And not using some nas software - too much overhead for what is access to nfs and iscsi.

Also, how will you back it up? Even if, somehow, a 10 GigE backbone gets setup, 20TB is still almost 24 hrs if not more. Adding in NAS layers means taking longer.

Having spent a few decades with hardware NAS's and HA based glusterfs, keep it simple and document how to recover when it blows up and all data disappears due to some raid glitch, os fault (lvm metadata corruption), ecc errors, multiple drive or controller port fail. It will happen.

@rsanders

I've a tape library. My plans were to save up some money to replace the LTO5 drive with a LTO 8 or 9 one. (9 once everyone switches to 10 and the used ones land on ebay).

@agowa338 Lto is pretty good stuff. We delt with proprietary backup software so no help there.

For my backups, rsnapshot has worked well. But that assumes a dedicated backup server. The one I had at work was fully automated and we never lost any data that had more than 6 hrs of age over the decade or so it was running. Includes replacing the backup server and drives.

For home, it's power the server on, sanpshot, power it off, except on Fridays for software updates before powering off.

@rsanders

Well the data doesn't change that frequently. It's more for "data hording" aka. archiving. So it is even unlikely for stuff to be changed or deleted. It's for the most part "amend only", so that kinda makes backing it up easier.

@rsanders

For back-ups I rely on a second server in a remote location. This is filling most of the same functions as my primary server - hosting movies, tv shows, and photos.

This second server is kept in sync with Syncthing. Although not an actual backup, the way Syncthing is configured I can copy a file back if I accidentally delete it on one end.

For the actual backup I have a QNAP DAS. When plugged into the NAS about once a month the NAS runs a backup script.

@agowa338

@agowa338

Been using Unraid for about 12 years now. I have about 28TBs of data.

I have experienced several failed drives - especially in the early days when I was "using up" a bunch of old disparate drives. I have never lost any data.

@agowa338 same here, been using unRAID for the last 8 years. It may fall short for some kind of black magic voodoo homelab usage I see here on my Mastodon feed, but overall it's a damn solid solution.
Hosting multiple services via docker for work, and extensive storage. Dealt with two disk failures, and recovered flawlessly!

@agowa338
I like it and have used it for years. I like the ability to use mismatched disk sizes, I can store what I choose on whatever disk, or let it spread data across disks. I like that if I have a failed disk, I can move stuff around and remove the failed disk without having to go buy another disk, then slap in a new one later, if I want.

In general, performance isn't great. Reading is same as 1 normal disk, but writing is slower. This is because the parity needs to be updated in real time. That means either (a) reading the data on the disk before writing over it, or (b) reading the data on that spot on all other disks. Caching writes to an SSD (ideally mirror), and having them later move to the array is a common and recommended setup.

The one pitfall I'd point out is very poor write IO in a failed disk scenario. Unraid will emulate the failed disk, so it appears accessible like normal for read and write. Every time you read a file on the failed disk, it reads that spot on all other disks and the parity disk to reconstruct what data was on the failed disk. If you want to move that data to a different disk, it needs to use all disks to reconstruct the data, then normal "read before writing" + actual write on the other disk + parity update. That is a lot of reading, and it slows everything down. It works, and I've needed it several times, but the recovery feels so slow.

@agowa338
Also, running VMs on an encrypted mirrored SSD cache has some write amplification, but would be the same on any system: https://random.blackpacket.net/2021/01/09/unraid-cache-pool-write-amplification.html
UnRAID Cache Pool Write Amplication

Beware of using the combination of a multi-disk cache pool in UnRAID and encryption. It results in a significant write amplication.