ZFS Raid10 on (2x2TB NVMe + 2x960GB NVMe) - is it possible / makes sense?

[gag]

Hello!
I got a new Hetzner AX61-NVMe server with 2x(1.8T SAMSUNG MZQLB1T9HAJR-00007 NVMe drives) + 2x(894.3G KXD51RUE960G TOSHIBA NVMe drives) for the Proxmox VE single node server.
The server will host 5-6 virtual machines for websites (Centos7 + Bitrix Site Manager, Centos6 + LAMP). Most likely they will use KVM (LXC is being considered as an option).
File system usage for the biggest VM is ~ 300GB, MariaDB database size is ~ 20GB.

My questions are:
1) Is it possible, besides with boot, system and backup partitions on 1.8TB drives (mdadm RAID1 + LVM), use 2x800GB partitions on these drives along with 2x800GB partitions on 894.3G TOSHIBA NVMe drives to create a ZFS striped/mirrored pool (RAID10)?
2) If possible, would this be the right decision in terms of VMs performance, reliability and maintenance simplicity compared to the option of distributing the overall disk load between two ''classical" LVM on software RAID1 groups: first of them on two SAMSUNG NVMe, and second on two TOSHIBA NVMe?
3) Than again, if creating ZFS (RAID10) from different disks is available and makes sense, what would be the best configuration for it?
---
Sorry, English is not my native language, so please be kind to my mistakes )

 

[wolfgang]

Hi

1) Is it possible, besides with boot, system and backup partitions on 1.8TB drives (mdadm RAID1 + LVM), use 2x800GB partitions on these drives along with 2x800GB partitions on 894.3G TOSHIBA NVMe drives to create a ZFS striped/mirrored pool (RAID10)?

yes, it is possible, but it is not recommended. I never test it with NVMe and they are normally good with parallel disk access. The problem with normals Disks (SATA/SAS) is the performance impacts.
Also is this setup more complex and what is the goal?
I would make two different ZFS mirrors (rpool, data). This is very easy to manage and you use the performance different of the different Disks.
Because if you make a Raid10 all disk will be as fast as the slowest disk.

 

[gag]

Thank you so much, Wolfgang!
You've fully confirmed my assumptions that creating a single ZFS aray an all four NVMe disks in that particular case does not make any practical sense.