Avoid SMR drives and RAID cards in ZFS homelabs
Learn why hardware RAID controllers and SMR hard drives destroy ZFS performance and reliability, and how to use HBAs, CMR disks, and persistent names instead.

Most new homelab builders focus on CPU benchmarks or network throughput, assuming that raw disk speed is the final bottleneck. That is the trap. A pool built on a hardware RAID volume, or on SMR drives bought because they were the biggest for the money, using /dev/sda-style names, works for months. Then a disk fails, the RAID card reports "optimal" while ZFS sees checksum errors it cannot repair, the resilver onto the new drive crawls for days, and after a reboot the pool imports with a different disk in slot sdb.
The book Store It Right opens its first chapter on hardware with exactly this warning. This post covers the three critical hardware decisions that determine whether your ZFS pool survives long-term: choosing the right controller interface, selecting the correct drive technology, and naming disks persistently.
The trap: trusting a hardware RAID card
A hardware RAID controller presents a virtual disk to the operating system. ZFS then sees one big "drive" and cannot tell which physical disk returned bad data, cannot repair it from parity it does not control, and cannot read SMART data through most RAID firmware. The card's write cache can also acknowledge writes that are not on disk yet, which undoes ZFS's careful write ordering.
The book explains that a RAID card hides the disks behind one virtual volume, so ZFS can detect corruption but not fix it. An HBA in IT mode passes every disk straight through, so ZFS sees each drive, its serial number and SMART data. What ZFS wants is an HBA (host bus adapter): a controller that simply connects disks. The common homelab choice is a used LSI/Broadcom SAS card of the 9300 series class running IT mode (initiator-target) firmware, which turns off all RAID logic.
Many cards sold for servers ship in IR (RAID) mode; some can be cross-flashed to IT firmware. The procedure differs by card and firmware generation and can brick a card if done wrong, so check your card's documentation and the vendor's firmware notes before flashing anything.
WATCH OUT "JBOD mode" or "passthrough mode" on a RAID card is not the same as IT mode. Some cards still hide SMART data or reorder writes in those modes. If you cannot see each disk's serial number with smartctl -i (Chapter 7), ZFS cannot see the disk properly either.
Motherboard SATA ports are fine for ZFS when the pool lives on the host itself. You only need a separate HBA when you run out of ports or when you want to pass all the disks to a TrueNAS VM in one piece.
The trap: buying SMR drives for capacity
Hard drives record data in two ways. CMR (conventional magnetic recording) writes tracks side by side. SMR (shingled magnetic recording) overlaps tracks like roof shingles to fit more data, which means rewriting one track forces the drive to rewrite its neighbours. SMR drives hide this behind a cache, which is fine for occasional writes and terrible for long sustained ones.
CMR tracks sit side by side and can be rewritten one at a time. SMR tracks overlap, so a sustained rewrite (a resilver, a scrub repair, a big copy) fills the drive's cache and throughput collapses. A resilver is exactly that workload: hours of sustained writes to one disk. SMR drives can slow a resilver from hours to days, and a long resilver is the window in which a second failure loses data. Use CMR drives for every ZFS pool. Manufacturers list the recording technology in their datasheets; check the exact model number, because the same product family can include both.
The book provides a clear guide for interface selection:
| Interface | Good for | Watch out for |
|---|
| SAS HDD | The same, with dual ports and better error reporting | Needs a SAS HBA or backplane; often louder |
| SATA SSD | Boot mirror, small fast pools, special vdevs | Consumer drives without power-loss protection |
| NVMe SSD | VM disks, databases, a fast mirror pool | Heat; limited M.2 slots; PCIe lanes on small boards |
A typical homelab NAS ends up with two pools: a mirror of SSDs or NVMe drives for VM disks and containers, and a pool of large hard drives for everything else. Keep them separate; mixing fast and slow disks in one pool makes it as slow as its slowest vdev.
The trap: using /dev/sda names
Linux hands out /dev/sda, /dev/sdb in the order disks are detected, and that order can change when you add a controller or a disk is slow to spin up. Persistent names live in /dev/disk/by-id/ and include the model and serial number, which is also what is printed on the drive's label.
The book advises building every pool with these names. When a disk fails, zpool status then tells you the serial number of the drive to pull, and you never have to guess which sdX it was. You can verify your disk IDs using the following commands:
ls -l /dev/disk/by-id/ | grep -v part
# ata-WDC_WD80EFPX-68C4ZN0_WD-RD1ABC12 -> ../../sda
# ata-WDC_WD80EFPX-68C4ZN0_WD-RD1ABC34 -> ../../sdb
# nvme-Samsung_SSD_990_PRO_2TB_S7DNNJ0X123456A -> ../../nvme0n1
lsblk -o NAME,SIZE,MODEL,SERIAL,ROTA
TIP Write the last four characters of each serial number on a label stuck to the front of the drive bay. Ten seconds of labelling now saves pulling the wrong disk from a degraded pool later.
Passing an HBA through to a VM
For the TrueNAS-in-a-VM design, the VM must own the disk controller outright. Proxmox does this with PCI(e) passthrough, documented in the Proxmox VE admin guide. The steps, in order:
- Enable IOMMU (Intel VT-d or AMD-Vi) and virtualization in the firmware setup.
- On AMD CPUs IOMMU is on by default; on Intel it is on by default with kernel 6.8 or newer, which is what Proxmox VE 9 ships. Older kernels need
intel_iommu=onon the kernel command line. - Load the VFIO modules at boot and refresh the initramfs.
- Check that the HBA sits in its own IOMMU group.
- Add the device to the VM.
# Confirm the IOMMU is on
dmesg | grep -e DMAR -e IOMMU
# Load VFIO at boot
printf 'vfio\nvfio_iommu_type1\nvfio_pci\n' > /etc/modules-load.d/vfio.conf
update-initramfs -u -k all
reboot
# After the reboot
lsmod | grep vfio
lspci -nn | grep -i -e LSI -e SAS
pvesh get /nodes/$(hostname)/hardware/pci --pci-class-blacklist ""
Checklist
- Replace any hardware RAID controller with an LSI/Broadcom SAS card of the 9300 series class running IT mode firmware, ensuring it is flashed correctly according to vendor documentation.
- Verify that your HBA is in true IT mode and not a "JBOD" or "passthrough" mode that still hides SMART data or reorders writes.
- Select only CMR drives for bulk storage pools, checking datasheets for the exact model number to avoid SMR variants that collapse during resilvers.
- Use persistent names from
/dev/disk/by-id/when creating ZFS pools, never relying on/dev/sdXdevice nodes that can shift after reboots. - Label the last four characters of each drive's serial number on the physical bay for quick identification during failures.
- Enable IOMMU in firmware and load VFIO modules at boot if passing an HBA through to a TrueNAS VM via PCI passthrough.
- Ensure the HBA resides in its own IOMMU group before adding it to the VM, avoiding ACS overrides unless absolutely necessary.
- Export any ZFS pools on the host before starting the VM that owns the HBA, preventing simultaneous access and potential corruption.
