Hi All,
Recently our small MSP has decided to move away from ESXI and move to Proxmox for our client server deployments. As I am heading up new server projects, I am going through learning this new environment and need some direction as to what the proper setup should be for our use case.
Before we made this decision to switch over, our few latest server deployments were going to use ESXI with x3 4TB drives in a RAID-5 running a singular Windows Server VM.
After doing "enough" research, I decided to install Proxmox on the new servers in RAID Z1 thinking this would be the same as RAID-5 and installed 1 windows server VM using a guide.
I then happened upon this post [here] that may indicate I may be doing something wrong.
If anyone has some basic configurations I could reference for this use case or direct me to some other guides I would really appreciate it.
You’re not far off, but there are a couple of conceptual differences that are important here.
RAID-Z1 in ZFS is often compared to RAID-5, but it’s not a 1:1 equivalent. It gives you single-disk redundancy like RAID-5, but the way ZFS handles writes, checksumming, and caching is fundamentally different. That’s where most of the “gotchas” come from when people switch from ESXi-style thinking.
For a setup like yours (3×4TB, single Windows VM), the main question is actually less about “what RAID level matches RAID-5” and more about whether ZFS RAID-Z1 is the right baseline for production workloads on small arrays.
A few practical points for Proxmox in MSP-style deployments:
With only 3 disks, RAID-Z1 works, but your usable performance and rebuild safety margins are not great. One disk failure + resilver stress is a real risk window.
Many Proxmox setups for small clusters actually prefer mirrors (RAID1-like vdevs) over RAID-Z for VM storage, because random I/O performance is better and rebuilds are faster and less risky.
If you stick with ZFS, make sure you’re not overcommitting RAM and that ARC sizing is reasonable for the host workload.
For Windows Server VMs, disk latency matters more than raw sequential throughput — this is where mirrors often outperform RAID-Z1 in practice.
So your instinct wasn’t wrong, but the assumption that RAID-Z1 = RAID-5 in behavior and suitability is where things diverge.
If this is going into client deployments, I’d strongly recommend standardising a few reference architectures (e.g. “2-disk mirror per node for small systems”, “ZFS RAID-Z2 for larger arrays”, etc.) rather than treating it as a direct ESXi RAID replacement.
