Creating a VM
Every choice the create form asks for, and which ones you cannot change later.
VMs → Create. The form is long because a virtual machine has a lot of properties; most have sensible defaults and only a handful genuinely need a decision.
Identity and placement
| Field | Notes |
|---|---|
| Name | Must be unique |
| Availability zone or host | One or the other, never both |
| Tags, notes | Free-form, for your own organisation |
In a deployment with zones, pick the zone and let the scheduler place the guest. Pick a host only where no zones are defined. See Availability zones.
Note — If the zone has to move or stop other guests to fit this one in, the form tells you which ones before you confirm. That consent is part of the request; the platform will not reshuffle other people’s workloads behind your back.
CPU and memory
| Field | Notes |
|---|---|
| vCPUs | Virtual CPUs |
| Memory | In MB |
| Sockets / cores / threads | Optional explicit topology |
Leave the topology alone unless the guest’s licensing or NUMA behaviour depends on it. Some operating systems and databases are licensed per socket, and presenting 8 sockets rather than 1 socket × 8 cores can be expensive.
Hugepages reduces memory-management overhead for large guests. It requires hugepages to be configured on the host and reserves memory up front.
Firmware
| Field | Notes |
|---|---|
| Firmware | BIOS or UEFI |
| Machine type | The emulated chipset |
| Secure Boot | UEFI only |
Warning — Firmware is effectively permanent. An operating system installed under BIOS will not boot when switched to UEFI. Decide before you install the guest, not after. Modern guests want UEFI; Secure Boot on top of it if the guest supports it.
Disks
Add one or more. Per disk:
| Field | Notes |
|---|---|
| Size | In GB |
| Backing | Storage pool, iSCSI LUN, or replicated storage image |
| Format | Disk image format |
| Bus | virtio unless the guest cannot |
| Cache mode | Leave at the default unless you know why |
| Bootable | Which disk the firmware boots |
Backing decides migration. A disk on a host’s storage pool is local: the VM cannot move to another host until the disk moves. A disk on an iSCSI LUN or a replicated storage image is shared, and the VM can migrate freely.
Choose shared backing for anything you expect to keep running through a host failure.
Use virtio for the bus. Other buses exist for guests whose installer has no virtio driver —
older Windows, chiefly — and cost performance.
Network
Add one or more NICs. Per NIC:
| Field | Notes |
|---|---|
| Attachment | A distributed switch, or a host bridge and port group |
| Model | virtio unless the guest cannot |
| MAC address | Generated; override only if something depends on it |
| Queues | Multi-queue for high-throughput guests |
| Rate limit | Cap in Mbps |
Attach to a distributed switch to get a network that follows the VM across hosts. A host bridge ties the VM to that host’s physical layout. See Distributed switches.
Installation media
Attach an ISO from the ISO library to install from. The library is shared across the cluster, so the same ISO is available wherever the VM is placed.
You can also start from a template, which skips installation entirely.
PCI passthrough
Assign a physical device — a GPU, an HBA, a NIC — directly to the guest. The form lists devices the host has available.
Warning — A VM with a passthrough device is bound to the host holding that device. It cannot live-migrate, and it cannot be restarted elsewhere by HA. Passthrough buys performance at the cost of mobility.
Availability options
| Field | Notes |
|---|---|
| HA enabled | Restart elsewhere if the host fails |
| HA priority | Restart order among HA VMs |
| Anti-affinity group | Keep apart from others in the same group |
| Evictable | May be powered off to make room for something more important |
| Auto-start | Start when the host boots |
| DR capable | Replicate to the paired site |
DR capable is applied at the end of the build, once the disks exist. See Disaster recovery.
What happens next
Creating a VM returns immediately and does the work as a task: disks are provisioned, the network is attached, the domain is defined. Watch the task for progress; a failure there carries the reason.
If auto-start is off — the default — the VM is created stopped. Start it and open its console to install the guest operating system.
What you cannot change later
Most things are editable. These are painful or impossible:
- Firmware type — reinstall the guest.
- Machine type — changing it can make an installed guest unbootable.
- Boot disk bus — the guest needs a driver for whatever it boots from.
Everything else — CPU, memory, extra disks, NICs, tags, HA settings — can be changed afterwards.