Thin or not to thin... LVM

[ixproxmox]

What would be your best advice? Setting up new server, that will have 2x256 GB SSD (OS in HW RAID1) and 4x16 TB (Data/VMs - HW RAID5). All datacenter rated disks and very good hw raid-card that will support all disks. 10 Gbps network.

For the main vm on this host, it will be running R1Soft CDP backup (emergency backup node, not primary CDP-backup, so it will not run that often as a normal backup) and it will use 30% of the disk capasity from start (and 60% 3 years ahead).

For the rest, it will be running different smaller vms for "fun" and other use cases where storage is needed.

CDP is a bit write intensive, so for this use, I would by default go for LVM and not thin mode. But at same time, it is tempting to run multiple services and the option to overcommit on these additional vms without having to create a seperate hw raid partition. For instance, I migth run FreeNAS and a 500 GB local disk, but only use 100 GB on it. Then it feels a bit waste to allocate 500 GB...

Any tip? I'm afraid of slow disk-operations for the CDP backup (at night), in a perfect would I would have lvm-thin if it has only like 5% performance hit, but reading posts i see anywhere up to maybe 40% performance hit.

 

[aaron]

You could also manually create a thick LVM and parallel a thin LVM and add PVE Storages for both. Then you could split the disk a bit and use thin provisioning for some of them.

If you can use an HBA, then you could consider using ZFS which makes it easier to have a storage configured that does not thin provision, and one that does on the same pool, located in different datasets. They both would share the total available space of the pool.

 

[ixproxmox]

You could also manually create a thick LVM and parallel a thin LVM and add PVE Storages for both. Then you could split the disk a bit and use thin provisioning for some of them.

If you can use an HBA, then you could consider using ZFS which makes it easier to have a storage configured that does not thin provision, and one that does on the same pool, located in different datasets. They both would share the total available space of the pool.

You mean create two volumes (basically /dev/sdb and /dev/sdc) from the hw raid controller and then create a thick lvm for one (sdb) and thin for (sdc) from inside Proxmox. That should work, only is if I need to increase the sdb at once time, not that easy from the raid-controller I think. In that sense maybe easier with zfs, but then I would kind of loose (some of) the point of hw raid perhaps..

I have an issue now that seems to indicate that proxmox doesn't like so big vms as 35TB...

 

[aaron]

No, I mean, on the disk, once you manually created a volume group (VG), you can add create a thin volume on top/next to it and define the additional storage there.
So you have two Storages on the same disk, one using regular LVM, one using the LVM thin pool. I did not test this myself and cannot give you the exact commands, only the rough idea of what should be possible.

Alternatively, I can recommend ZFS which makes handling this quite a bit easier. It doesn't like HW raid though

 

[ixproxmox]

No, I mean, on the disk, once you manually created a volume group (VG), you can add create a thin volume on top/next to it and define the additional storage there.
So you have two Storages on the same disk, one using regular LVM, one using the LVM thin pool. I did not test this myself and cannot give you the exact commands, only the rough idea of what should be possible.

Alternatively, I can recommend ZFS which makes handling this quite a bit easier. It doesn't like HW raid though

When I try to create a thin-volume, it says that I have no available disk But you are saying it is possible in theory to create a lvm-thin from the command line on the proxmox-host instead of using the GUI, I assume?