Question on PBS setup and hardware

[marar]

I have ran PBS in the past on bare metal and had no issues, but my setup has since change and looking for some help on which direction I should go if not running bare metal anymore.

Would it be better to ...

1. Install PBS as VM on Proxmox and backup to a NFS share on Synolgoy.
2. Install PBS as VM on Synology.

 

[UdoB]

1. Install PBS as VM on Proxmox and backup to a NFS share on Synolgoy.
2. Install PBS as VM on Synology.

Both (!) is suboptimal. (You know that, right?)

But if I am forced to choose: the PBS (or any backup system) should be independent from the source it will backup. You'll probably want the backup to be available and usable when (not: if!) the source dies, right?

 

[DerekG]

Would it be better to ...

1. Install PBS as VM on Proxmox and backup to a NFS share on Synolgoy.
2. Install PBS as VM on Synology.

I suggest that option 2 is preferable, except it won't protect you in the case of fire or flood in the home, but I'm sure that you are aware of that.

 

[marar]

Both (!) is suboptimal. (You know that, right?)

But if I am forced to choose: the PBS (or any backup system) should be independent from the source it will backup. You'll probably want the backup to be available and usable when (not: if!) the source dies, right?

Technically both options the backup is still available if the host dies, though one is a little harder to recover. I know it's not optimal and is why i have ran bare metal up until now. Just looking at best option short term and know many people run it in a vm on pve.

 

[marar]

I suggest that option 2 is preferable, except it won't protect you in the case of fire or flood in the home, but I'm sure that you are aware of that.

Thank you, yeah fire and flood would not be good. I do have cold storage off site for more important things to try and help with that.

 

[Eduardo Taboada]

You can try to use the NFS as a removable storage in PBS, if the PBS host dies, you can create a new one, and add the storage again as a removable existent media.
If you use this trick you can access again to your backups.

 

[DerekG]

And a ZFS Pool with 2x VDEV[0,1] zfs mirror of 2x HDDs and VDEV2 zfs special device of N x SSDs N >1 can/ will work too.
I give my proxmox bs with zfs, 32 gbyte ram and up to 24 gbyte ram arc cache.

I understand that those are best practices, but those specs are difficult or too costly for most who are building a homelab, remember that for most it's just a hobby or a learning tool, so the costs matter.

 

[floh8]

Use the 2. solution with btrfs raid10 and xfs fs with 64 kB block size.

 

[Johannes S]

I would prefer the second one since with that you don't need a network connection for the verify and garbage-collection jobs. The German forum had a thread on it: https://forum.proxmox.com/threads/warum-ist-der-pbs-so-picky-bei-nfs-storage-mounts.160610/

Basically the user switched from option 1 (PBS VM on PVE host, NFS share on NAS as datastore) to option 2 (PBS VM on the NAS) which ended in a huge speedup for the garbage-collection and verify jobs.

If your budget allows a seperate used Mini-PC with just PBS installed is another option for a low-budget backup server.

 

[alexskysilk]

Both (!) is suboptimal. (You know that, right?)

why? having PBS and its payload storage integrated into one device seems like a perfectly reasonable solution... Agreed that option 1 is a non starter.

 

[UdoB]

having PBS and its payload storage integrated into one device seems like a perfectly reasonable solution...

Yes, it works.

I said "suboptimal". For me this means that some potential is wasted - for one or another reason. In this case we know that PBS wants IOPS. For me this means that the best results require direct hardware access to fast storage, e.g. SSDs. Each abstraction layer like virtual disks for the chunk-store or traversing a complex network stack to reach out for NFS adds latency.

There are a lot of different ways to implement a backup system. And a "suboptimal" PBS is much, much better than no backup

 

[alexskysilk]

For me this means that some potential is wasted - for one or another reason. In this case we know that PBS wants IOPS

This is completely backwards. PBS doesnt want anything. load is generated by the sender, and is limited by the host storage and network connection. without establishing what your load is, you're not "wasting" anything.

More to the point, the payload storage is already established; you're not going to make it substantively faster by having it accessed by a standalone host as opposed to a local vm.