Hello, I have been watching on YouTube for years and finally decided to sign up for the forum.
I have been running in circles trying to figure out a “best/optimal” way of building/deploying a PBS.
My options that I have:
1: Terra Master D8 Hybrid hooked into a HP mini PC, 7500T with 16gb of RAM using 4x 8tb drives
2: (current deployment) VM in Unraid with HBA pass through to the VM with the 4x 8tb drives
3: A VM on my Proxmox Server, AMD 3960X, and use the Terra Master to keep the 4x 8tb Drives and the OS drives “separate” from the server it is backing up. (I know this isn’t the best per se, but it would eliminate a system in the rack, and put to use the compute and RAM in the server more)
4: Run a PBS docker on my Unraid server and use the SSD array for backup (after the rebuild I should have 16tb of space in raidz2)
5: Run a VM on the Proxmox server and use network storage for the datastore (I know most stuff online say to not do this, but I’m not production, just a hobby homelab)
I’m open to any advice and suggestions. I’m excited to post here more in the future, learn, and help when I can.
Additional info: I don’t run too many VMs, it is just a home lab as a hobby. I probably only need 4-8tb of backup now, but figure it would be a pain to expand later. I have mostly Ubiqiti networking with 10g readily available.
Figure out how to apply the 3-2-1 backup rule to your homelab.
I have a PBS in LXC container on the always-on server running PVE for frequent (every 2 hours) backups.
This protects mostly again human error (“oh, I shouldn’t have clicked this button!”), also against failed sw updates. It offers a true copy, but on the same storage, so no protection against hw failure.
Then I have a PBS on separate dedicated hw pulling (sync) from LCX on a daily basis.
The physical instance is turned off most of the day and has read access to the PBS LCX instance, but no reverse access.
This is a true second copy running on separate hw. Saved my bacon a couple of times.
This configuration is “almost” airgapped with the intention to limit horizontal migration in case of a breach (getting hacked).
I have a third PC configured in the same way as #2 with the intention to store it at a friends house.
This is to protect against natural disasters and other catastrophic events (theft?).
I still need to work out kinks to instill the trust that I can leave this PBS running unattended for an extended length of time.
Not saying this is what you should do, but I offer it as a data point to kickstart the process.
A lot of good points and some things I have considered and others I have not.
So I ended up playing around with some stuff and I think i found a solution…. maybe an unholy one.
I have PBS running in a docker on the Unraid box(figured it was better to keep the running server off the same box maybe? If this is just a silly thing to do I can run it on the Proxmox Server instead). I added a remote NFS share from my UnasPro 8 (That I will have indexing on to help prevent the human error you mentioned, along with having a elongated prune schedule on the PBS) I then added that share to the docker in the configuration. Loads right up and runs quite fast. I tried the same thing with SMB and that was a mistake…. painfully slow.
Now for the offsite, that was a separate but now related project. I have been building a second Unraid box, loaded tailscale onto it and a gli KVM that also has tailscale and a motherboard breakout. That way I don’t have to bug my friend to update a bios or turn back on if there was a power failure. I plan on adding this remote storage into the mix and running occasional backups to it as well. If I get frisky, I may try and load up another PBS on that Unraid box as well using local storage.
Not sure how much this is a decent solution or a ticking time bomb but I think it should do ok (please tell me if I overlooked something). I think that would do the 3-2-1. 3 copies, ssd and HDD, and an offsite.
This also eliminates a separate block of storage for a stand alone PBS and uses the storage in the UNAS (there is plenty of room and can easily expand). With drives so expensive, anything I can do to use less I want to. It also will keep one less system running and that means less electricity too.
Thank you for the suggestions, I’m going to keep some of that in mind. Also means That I’m not a total fool for not running the PBS on bare metal.
Well, 2026 is the year to experiment with software in your existing set of hardware that makes up your homelab.
If you want/need to expand storage and have some money burning in your pocket - just not enough for RAM and SSDs/HDDs - this is a great year to add long-term tape storage to your homelab.
I was fascinated how (relatively) cheaply you can get a tape library with LTO drive and a bunch of tapes.
As you probably saw, PBS supports tape storage ootb.
I have the tape library connected to the second server for another longterm copy.
Didn’t mention it in the first post because it is quite esoteric. Fun to explore but not the first thing to recommend.
I have never considered this an option, but I do remember seeing tape drives being supported.
I need to finish at least a few more of the dozen or so projects I’m playing with before I add something like that to the mix, but it does sound fun.
I miss the old Zip Drives for backing things up. You could fit dozens of floppies on one of those and put it somewhere safe.
Thanks again. After I get this stuff settled I want to try and Intel B60 with SR-IOV running 4x VMs going out via Moonlight to light weight systems for my kids and their friends to play games on this winter.
I’m blaming Wendell and his vid from a few days back for this latest idea. But it will free up some PCIe lanes in the Threadripper box and then I can do other fun things.
Storing the pbs backups on mechanical drives or even a nfs share with mechanical drives on a nas is perfectly fine for a homelab or even a small businness setup with like 10 vms.
If you read some places it gives you the impression that after a while it will spontaneously combust or grind down to unusable speed if you are not on bare metal, have heaps of RAM and should consider a special mirror for the local zfs array. (I may have snagged some small industrial SSDs on ebay to test with to see if this was needed)
If I get up to 10 VMs I will have found some additional use cases and be quite pleased with myself.
Thats probably going to help. The backups split the vm into a bunch of 4mb files (chunks), so even in a small setup you are dealing with hundreds of thousands of 4mb files. Similarly, any storage optimization should be around large numbers of 4mb files. So for example increase zfs recordsize from default.
So, the UNAS8pro that I have storing the backups has a 2x 512gb cache. It says it is in the high 90s for hitrate read/write. I’m going to guess that is helping, but the meta data is the bigger issue I take it.
A read and write hot-spot cache, using the dm-cache kernel
module. This cache tracks access patterns and adjusts its
content deliberately so that commonly used parts of the main LV
are likely to be found on the fast storage. LVM refers to this
using the LV type cache.
It works at the block level, so it has no concept of data or metadata in the filesystem, but its reasonable to assume that any metadata is probably in this cache because any activity is going to involve a lot of metadata lookups to see where each 4mb file is.
The difference with a zfs special device for metadata is the wear on the ssd. Updating metadata is not a whole lot of writing, while “simple” caching like in that nas will make it write also some of the data from the backups in the cache as they land on the nas
If I keep this setup…. I will have to keep an eye on the wear level on those SSDs. I did spend a bit extra to make sure they have DRAM cache. So they shouldn’t instantly incinerate.
In my view, keep PBS turned off isn’t such good idea, as it can’t perform verify and cleanup tasks, which I see critical.
Also, I’m encountering strange errors with containers backup, it takes full stop time after time.
But our topicstarter especially points to the “Passthrough disks” - I really don’t know how PBS will handle this, as snapshot mode is not an option, and turned off machine will not provide disks.
Actually, PVE is hardly tied to ZFS, and (something-something) NFS.
Look what they did with iSCSI and LVM - the pure pornography as I see it.
So for announced architecture PBS will not go.