Choosing Discs: Consistency vs. Off-Site Backup

[Des]

Hi,

I am new to FreeNas and ZFS and am considering to move there. I have experience running several servers at work. I am currently re-thinking my storage and backup strategy at home. It seems that I have to decide between Raid2 for my data or offsite backup for my data.

My situation:
- 6 Desktop/Laptop devices of family members that run backups to my NAS.
- Besides these Backups, the NAS stores data (documents, pictures, videos, etc)
- Total requirement is about 1TB at the moment. I expect it to 2tb for the next couple of years.

I have two WD red 3TB and one WD green 2TB (plus two other 1TB discs) available. I won't be able to spend more money the next time, so I am set with this.

Offsite-Backups are important for me. So far, I was running a one-disc nas (ext4) and running nightly rsyncs to a separate disc. I was rotating the separate disc offsite on a monthly basis.

That worked very well for me, but I was having trouble with silent corruption which I did notice by accident. I was able to recover, but anyway, I want to to address silent corruption, bit rot or whatever.

It seems that I have to decide between redundancy with raid2 and offsite. I was thinking to choose offsite:

A) Run the NAS with ZFS on a single disc only. Run nightly rsync or use zfs/send to get the data to a second disc. I will rotate this disc with a third one offsite. This will work as long as my data is below 2tb which is fine for the next years. I am aware of the fact that ZFS will not be able to heal corrupt data. I assume that I can get back the files from my backup (second built in or the older one offsite)? But I assume that I can get notified on curruption and that there is no "silent corruption"?

B) Run the NAS with ZFS in Raid2. Then I will not have discs left for rotating offsite. Just one to take stuff offsite. As redundancy is not my main concern, I was thinking not to go for this one.


Which option would you take and are there any arguments or trouble that I am not seeing with A)? At this moment, I am not able to spend extra money into a extra drive to have two discs for mirroring and two for rotating backups.


Thanks in advance!

 

[jde]

You need to post the hardware you intend to use with the disks. I believe a single disk vdev can detect bit rot, but cannot correct it. So, a single disk has won't help you there. You need redundancy for that. Also, If you are concerned about bit rot , you should also be concerned with having ECC memory. I get the sense that you are trying to put together a build with hardware you already have lying around. If the hardware doesn't support ECC memory, bit rot might be the least of your problems. https://forums.freenas.org/index.php?threads/ecc-vs-non-ecc-ram-and-zfs.15449/

Edit:
You may be able to get redundancy with a single disk by setting copies to 2. I don't have any experience doing so. Perhaps other members of the forum could give you info on doing so. Doing so will also cut your storage space in half.

 

[jgreco]

You need to post the hardware you intend to use with the disks. I believe a single disk vdev can detect bit rot, but cannot correct it. So, a single disk has won't help you there. You need redundancy for that. Also, If you are concerned about bit rot , you should also be concerned with having ECC memory. I get the sense that you are trying to put together a build with hardware you already have lying around. If the hardware doesn't support ECC memory, bit rot might be the least of your problems. https://forums.freenas.org/index.php?threads/ecc-vs-non-ecc-ram-and-zfs.15449/

Edit:
You may be able to get redundancy with a single disk by setting copies to 2. I don't have any experience doing so. Perhaps other members of the forum could give you info on doing so. Doing so will also cut your storage space in half.

This is entirely correct.

 

[Robert Trevellyan]

You could mirror your two 3TB Reds for redundant storage of your 1TB of live data (growing to 2TB while staying below 80% utilization), in your NAS. This setup will be able to detect and correct bit rot and can survive the failure of one disk.

Then you could rotate the most important data in offsite backups on other discs. If the offsite backup disks are single disk ZFS pools, you will at least be able to detect bit rot and make a new backup.

 

[Des]

Hi,

thanks a lot for your replies. I am planning to use a one-year old mainboard and upgrading to 8GB ECC memory.

My idea was to only use a single pool ZFS in the NAS to only detect problems, but not to correct them. I would correct them from by (offsite) backup. From what I understand, this would work. I currently cannot estimate the effort of replacing the corrupt parts with backups. Can anyone share some experience? I will know that file XY is corrupt and then can replace that file from backup?

I read about copies=2 for one disc setups with detection and correction but also read that this drops performance and of course available space (which is not yet a problem).

So, I take that you would recommend to have NAS redundancy over offsite backups. I agree with taking the most important data offsite on the on TB drives. Trouble is, that I cannot (at the moment) tell what is important data: I also have backups from the mentioned clients on the NAS but cannot tell what part of that backup is important to the family members. Maybe I have to work on that.

 

[jgreco]

You cannot turn off error correction in ZFS except by disabling error detection (i.e. checksums).

My observation over many years is that people often don't do anything to protect their data, or the steps they take are halfway at best. ZFS was designed to do the things that an admin OUGHT to do, such as monitoring for corrupt files, but which would normally be way too much work and effort, so almost no one does.

Why would you purposely want to have a disk read fail, and then have to manually restore files from an offsite backup, if the filesystem is smart enough to correct the error and make it such that anything that's reading the file is able to read it successfully?

 

[Robert Trevellyan]

I take that you would recommend to have NAS redundancy over offsite backups.

Not exactly. Redundancy improves data availability, but backups are still necessary for anything you care about. It only makes sense not to backup the contents of your NAS if one of the following is true:

1. They are of no value, or are easily replaceable.
2. They constitute backups of other devices, and you already have separate offsite backups.
3. They constitute backups of other devices, but they aren't important enough for offsite backups.

I agree with taking the most important data offsite on the on TB drives. Trouble is, that I cannot (at the moment) tell what is important data: I also have backups from the mentioned clients on the NAS but cannot tell what part of that backup is important to the family members. Maybe I have to work on that.

Emphatically, yes.