Looking for opinion / suggestions on a new PVE setup : OS on 2 SSDs, local backups on 2 SATA drives

[magicfab]

I am setting up a new PVE node as follows :
* 2 x 480 GB SSD for Proxmox VE (Z-RAID1)
* 2 x 2TB for local backups and other storage (RAID-1)

CPU will be Intel Xeon E-2334 @ 3.40GHz, 64GB of RAM.

The load would be about 12 VMs, mostly Debian 11 with light loads, but also a couple of Windows 10 Pro that were recently virtualized to make backups and resource allocation easier. "Other storage" may be daily backups of some VMs (not all) and ISOs.

This would be clustered with another similar server. For VMs requiring minimal downtime, iSCSI storage will be available. I am still not sure which shared storage to use to make live-migration possible with minimal delays.

I am looking for suggestions on improving this (minimal) 2 node cluster setup. Any ideas/comments are welcome.

 

[Dunuin]

I am looking for suggestions on improving this (minimal) 2 node cluster setup.

You need at least 3 hosts for a cluster. Either 3,5,7,... PVE nodes or 2,4,6,... PVE nodes + a host as a qdevice. Maybe you already know that, but I wanted to mention this, as I answered two threads today, where people wondered why their whole 2 node clusters weren't working anymore after shutting down one of the two nodes. They didn't realize that wehn shutting down one node the remaining node will stop working too, because of lost quorum and the split brain situation.

Ceph is also out with just 2 nodes. ZFS would work with 2 nodes but ZFS isn't a real shared storage. Its just local storage that gets replicated/synced every X minutes. Not that great for HA, as the nodes won't be perfectly synced. So when a node fails and HA starts the VM on another node, that node will start an outdated version of the VM so you might lose the most recent seconds/minutes of data.
A NFS share could work as a shared storage, but then you got a single point of failure, as a failing NFS server would make all PVE nodes useless.

 

[magicfab]

Thank you, there is a retired node which is also scheduled to be replaced, apologies for not making that clear.
I have no intention on using Ceph here, but I plan to learn more about thin provisioning and other storage types. I know for sure I don't want to use NFS again, I've had bad experiences with it in the past.