LVM-thin vs. ZFS - for safety and snapshotting

[Andrew Bienhaus]

Hi all,
Hope you can fogive the newb question, but I'd like a couple of opinions.
Been at this 30+ years, but new to the linux vm world, and admittedly to "intense linux". Usually have paid the bills in the windows world, but am expanding my horizons!

A have a good friend, who swears by his VM system and their ZFS snaphots.
Cryptovirus hits a client, he can roll back 30 mins before it hit, in about 30 seconds, as he puts it.

We talked about the idea that I have to use ZFS or LVM-thin to snapshot, and, it was his opinion that while that is a true statement, that the LVM-thin model, can build up performance issues the more times you snapshot, versus the zfs approach. He described the design in detail, and what he said made sense... just wondering if there is anything "out of date" in his assertions about the use of LVM-thin snapshotting, and performance.

Thanks,
Andrew

 

[Andrew Bienhaus]

Oh, and I already have raid and such in place, seaparately... so performance through ZFS vs. LVM-thin, is the question here.

And therefore for that matter - if my boot drive will be holding VMs as well (900gb SSD), should I have picked ZFS as the default file system at install time too? (I can install it again)

 

[Dunuin]

I think the biggest benefit of zfs is, that it checksums everything. And from time to time it will scrub all drives, recalculating checksums of all data and compares them to stored checksums. If it finds corrupted data (bit rot for example) and there is parity data or a mirror, it will automatically fix it so everything is healthy again.
And write operations are atomic, so its really unlikely that a pool/array degrades if a power outage occures.
And it can compress and/or deduplicate on block level.
And you can replicate from one sever to another using snapshots.

Biggest disadvantages:
-not easy to add new drives to a pool/array
-needs a lot of RAM
-only as stable as the hardware it runs on (ECC and enterprise grade SSDs recommended)

 

[Andrew Bienhaus]

Thanks... but in a VM world, the Snapshotting, is a definite benefit. So that's the thing I'm trying to work out for the long run...

 

[Dunuin]

Thanks... but in a VM world, the Snapshotting, is a definite benefit. So that's the thing I'm trying to work out for the long run...

Keep in mind that proxmox allows you only to rollback to the last snapshot if using ZFS. With LVM and qcow2 you can rollback to every snapshot.
In theory you can go back to every snapshot by creating a clone based on any snapshot but that is advanced stuff proxmox isn't using. But as soon as you do a real rollback everything after the date of the snapshot creation will be deleted.

 

[Andrew Bienhaus]

Oh?
is that a ProxMox limitation then? (he's using KVM and something else in his systems)

 

[Dunuin]

Thats because everything newer then the snapshot you want to roll back to will be deleted. So you can't switch back and forth. If you rollback thats a oneway road and you can't undo it. Thats why ZFS allows you to create clones based on snapshots so you can work with the clones without needing to rollback. But that is stuff you need to do manually using the CLI and is not implemented in proxmox.

 

[Andrew Bienhaus]

ok, so, multiple points in time through whatever process (whether CLI by hand or ProxMox itself) is workable on either, just one easier than the other.

ADD: Because yes, we talked about how he has had clients infested, where he then takes a clone, or has scheduled transmissions of ZFS snapshots to alternate servers, and can load them up to see first, where he should restore back to... did the infestation happen last night at 5? or did it not happen until 3am? (in which case, you'd want to snap back to 230am sort of idea, keeping all data that arrived between 5pm and 230am)

What about the assertion that LVM-thin (with qcwo2 I assume) and it's method of storing the snapshots builds and builds, and slowly degrades performance?

 

[donrusso]

Hi all, i have one M.2 nvme SSD 1TB on my motherboard. SATA HDD's are also present. So wich filesystem should i prefere "RAID0 ZFS" or "LVM-Thin" for my root host M.2 ?? Actually i run 3 GB RAM with PVE + 1GBRam ubuntusrv VM

 

[Andrew Bienhaus]

Sorry, I'm not going to be much help. I overwrote proxmox with the free Hyper-V 2019 server from Microsoft and haven't looked back.
Maybe someone else here can finish this thought?

Too much "secret sauce" at work here.

 

[donrusso]

Sorry, I'm not going to be much help. I overwrote proxmox with the free Hyper-V 2019 server from Microsoft and haven't looked back.
Maybe someone else here can finish this thought?

Too much "secret sauce" at work here.

 

[Andrew Bienhaus]

-LOL- Sorry Don, I meant no offence.
I spent a month (if not two?) trying different hypervisors, and this one looked like the best option, but I kept running into knowledge walls. (not so much here, elsewhere I mean, and on other linux/kvm topics)
"Yeah, you can do that, I do it, I wrote some custom scripts to make it happen..." ... and then silence.
That's what I mean by "secret sauce". Everyone seems to have their own way to do it.
the ZFS vs. LVM here was just the final thing... and another user messaged me privately, and said "Have you tried HyperV yet?", as I mentioned that I had tried VMWare too.
I now have 2 x 2019 servers running on it, my Dad's XP system I needed to keep running, a win10pro client, two Ubuntu servers, and even an Ubuntu desktop. (the last one, just for fun)
YMMV.