I currently have two PVE2.3 hosts in a cluster. A 3rd host running Ubuntu/ZFS serves ISO/images to the hosts. The backup server (2nd proxmox) also runs a small storage appliance (also ubuntu/zfs) with two 1TB drives passed through to it. The main ubuntu sends snapshots every 5 minutes to the backup, but that is not real-time replication. From reading here (and elsewhere but mostly here), what I want to do is this: Have each proxmox server with a zfs block device (zvol) mirrored using drbd in primary/primary mode. I would create a proxmox storage (lvm on local/shared), such that each VM disk is an LVM (as this works already). The main server would also have the HBA with the SAS disk JBOD array, so VM disk reads would be extremely fast, due to no network choke point. Writes will be slower due to drbd, but that is not so critical to me. I did a mockup of this using two proxmox VMs (with KVM acceleration disabled) and it all seems to work fine. I would be upgrading to drbd 8.3.13 using the thread discussion here: http://forum.proxmox.com/threads/9376-Bug-in-DRBD-causes-split-brain-already-patched-by-DRBD-devs. Does this all make sense? Thanks!