It used to be that after a while (varies from a couple of weeks to months) Time Machine would complain that the spare disk image on the NAS was corrupted and you'd have to start from scratch, which completely defeats the purpose of backup.
I would assume that'd work fine, because I run my main workstations and a server full time with my home folder and various data living off APFS or HFS formatted iSCSI targets served over 10 gigabit from a TrueNAS system. I haven't upgraded to macOS 13 yet since I tend to be a lot more conservative these days about Mac upgrades than I used to, but Xtend SAN at least still works beautifully under macOS 12. I really, REALLY hope Apple doesn't remove that capability without a native alternative (or at least officially sanctioned way to do the same thing). Of course in that specific scenario Time Machine is of limited value (and I don't use it) since why bother when you can just use ZFS snapshots on the NAS side instead. TM has always been too flaky for me to feel comfortable with anyway.
Still, could be powerful in some instances, particularly when combined with a hardware gadget setup. Which leads to a response to sibling thread:
>Asmod4n 3 hours ago: Can’t do full system restores from iSCSI
Linux has kernel support for USB gadgets that can emulate mass storage devices, and that means you can actually rig up a little cheap embedded board to translate iSCSI into something that will look to anything else like a standard external USB hard drive. See for example [0]. Like the author there I use it to work around issues with appliances like consoles or the like that don't know how to natively deal with a lot of network stuff, but it's a pretty powerful tool if you've got a good NAS and LAN setup regardless. Get your full network storage flexibility, snapshots, data integrity and so on that pretty much anything ever with a USB port can utilize natively.
----
So far I've found it to be pretty reliable.
I've given up on Time Machine over SMB and I'm using Borg + Vorta for my backups. It has the benefit of being readable by any other Linux machine, too.
I'm not sure what causes TM to start considering the volume corrupted, because I believe at one point I switched over to a hardwired connection, but still saw a corrupted volume.
I compared that to Borg, which doesn't flinch if you SIGKILL the process mid-backup, and will basically just pick up from where it left off if you start another one. I saw no reason to continue with TM at all.
The pool itself is 8x12TB Toshiba enterprise drives. I'll have to check what pool config I used but I did add multiple NVME SSDs for SLOG and ARC. Whichever one is the important one is actually mirrored across the NVME drives for redundancy while the other partition only exists on one NVME drive, which was mildly annoying to configure.
I'm still not sure I made the "best" choices for the ZFS config - there are many knobs to turn.
Connections are a mix of wired and wireless. The NAS box has 10GbE to my UDMPro.
See my other comment for how the datasets are setup. I forgot to mention that I applied a quota to force each client to prune its history to avoid consuming all disk space.
Thankfully this is not my first rodeo. My Time Machine over SMB on ZFS setup has zfs-auto-snapshot set up, so I was able to roll back to a snapshot a month or so prior, which Time Machine happily verified, and it resumed backups without a hitch (keeping several years worth of backups intact).
Were there any specific tweaks you made to your setup to make this reliable?
I also wonder if the amount of backup data makes a difference. I have tended to use Time Machine to backup pretty much my whole drive. I wonder if it's more reliable when backing up smaller amounts of data perhaps...
As to amount of data, I run it on my whole drive with a few excluded directories, that's 700-ish gb of non-excluded data and the backup on the Synology is 1.6TB. I've recently migrated the TimeMachine backup from an older Intel MBP to a new M2 Air (I wasn't even aware you could do that) and that seems to have worked just fine as well.
I also found increasing the band size to 50MB from the default of 8MB made things work “better”, but I lack concrete empirical evidence.
In terms of stability, I haven't had a problem with TM on a NAS for a few years, but yeah the corruption is why I used to use CCC instead.
I'm using ZFS + netatalk in Gentoo server since 2015 as a backup server for 5 macOS devices without problems.
Early on I had an issue with over provisioning the ZFS slice or whatever but that was resolved.
FWIW, I've done Time Machine backups over Wi-Fi to a QNAP NAS for several years and have never run into this issue. I'm just using default settings for everything.