Learn NAS - Virtualization on NAS

#Introduction

Besides containers, a modern NAS can run full virtual machines. Episode 19 covers virtualization on the NAS: creating KVM and QEMU-based VMs on TrueNAS SCALE, understanding nested virtualization, passing a GPU through to a VM, and designing VM storage that's safe and easy to back up.

Virtualization gives you the freedom to run different operating systems on top of the NAS — from a virtual router, Windows for specific needs, to additional Linux servers. Because everything runs on top of ZFS storage, VM data enjoys the checksums, snapshots, and replication you've already built.

By the end of this episode you'll be able to create VMs on TrueNAS SCALE, understand the limits of nested virtualization, do GPU passthrough, choose between zvols and disk passthrough, and back up VMs with consistent snapshots.

#VMs on TrueNAS SCALE

#KVM and QEMU

TrueNAS SCALE uses KVM (kernel-based virtual machine) together with QEMU as the emulator. KVM leverages hardware CPU virtualization so VMs run near native speed. The Virtualization menu in the Web UI guides VM creation.

egrep -c '(vmx|svm)' /proc/cpuinfo

The command above counts cores with the Intel (vmx) or AMD (svm) virtualization flag. A zero means VMs must run in emulation mode, which is much slower.

#Creating Your First VM

When creating a VM, you choose the OS type, CPU, RAM, and ISO installation source. For storage, choose a zvol as the disk — this ties the VM to the already-protected ZFS storage.

midclt call vm.query | python3 -m json.tool

The midclt call vm.query command lists VMs along with their status. Use it to verify a VM is running and monitor the resources it uses.

#Nested Virtualization

#Virtualization Inside a VM

Nested virtualization allows a VM to run a hypervisor inside it — for example, running Proxmox inside a TrueNAS VM. This feature is useful for labs, but only if the host CPU and hypervisor support the nested flags.

modprobe kvm_intel nested=1

The modprobe kvm_intel nested=1 command enables nested virtualization on the host kernel. The module name follows the CPU vendor: kvm_intel for Intel, kvm_amd for AMD.

#Limits of Nested

Nested virtualization adds a layer that lowers performance and makes troubleshooting harder. Use it for labs and testing, not production. Also make sure the nested parameter is enabled persistently through the modprobe configuration.

#GPU Passthrough

#Why GPU Passthrough

GPU passthrough hands a physical GPU device directly to a single VM, giving the VM full access for workloads like media transcoding or rendering. On TrueNAS SCALE, GPUs are configured through the VM settings, and PCIe passthrough requires IOMMU support.

dmesg | grep -i iommu

The dmesg output shows whether IOMMU is active. Without IOMMU, PCIe passthrough isn't possible; enable it via the intel_iommu=on or amd_iommu=on kernel parameters.

#GPU Considerations

A single GPU can only be given to one VM. If the NAS serves many purposes, consider whether the GPU is truly needed in a VM or if it's better used by the host. Passthrough also disables the GPU for the host, so make sure no host service needs it.

#VM Storage

#Zvols for VM Disks

A zvol is the best choice for a VM disk: a block device backed by ZFS, getting checksums and snapshots. Because it behaves like a real disk, the filesystem inside the VM works normally.

zfs create -V 50g -o sync=always tank/vms/vm-01

The zfs create -V 50g -o sync=always tank/vms/vm-01 command creates a 50-gigabyte virtual disk. The sync=always option ensures a write is considered complete only after it's stored on disk.

#Disk Passthrough

Besides zvols, TrueNAS SCALE supports disk passthrough: handing an entire physical disk to a VM, for example for direct storage or data you don't want under ZFS control. The trade-off: the disk loses ZFS protection and can't be snapshotted as a single unit.

#VM Backup with Snapshots

#ZFS Snapshots for VMs

Snapshotting a zvol while a VM is running can produce an inconsistent state because the filesystem cache inside the VM hasn't been written to disk. The solution: combine ZFS snapshots with quiescing — for example, snapshot the VM from inside its OS before taking the ZFS snapshot.

zfs snapshot tank/vms/vm-01@sebelum-update

The zfs snapshot tank/vms/vm-01@sebelum-update command snapshots the VM disk. For full consistency, take the snapshot from inside the VM or pause the VM briefly while the snapshot is taken.

#VM Backup Strategies

• Regular ZFS snapshots for fast rollback.
• Replicate the zvol to a second NAS for offsite backup.
• Combine with file-level backups for critical applications.

Because zvols can be replicated like regular datasets, a whole VM can be duplicated elsewhere with zfs send.

#Closing

In this episode 19 you added virtualization to the NAS: creating KVM and QEMU-based VMs on TrueNAS SCALE, understanding nested virtualization, passing a GPU through, choosing zvol or disk passthrough storage, and backing up VMs with consistent snapshots.

Key takeaways:

• KVM and QEMU give VMs near-native performance.
• Nested virtualization is for labs only, not production.
• GPU passthrough requires IOMMU and serves only one VM.
• A zvol is the ideal choice for a VM disk with ZFS protection.
• VM snapshots must be combined with quiescing to stay consistent.

In the next episode, episode 20, we'll cover high availability and scale-out — from enterprise TrueNAS HA with failover and fencing, to scale-out with clusters, the MinIO and Garage object stores, and Ceph integration for distributed storage. One NAS is already great; now it's time to build a system that never stops.