I often get asked questions about using my software, particularly sway, on hypervisors like VirtualBox and VMWare, as well as for general advice on which hypervisor to choose. My answer is always the same: qemu. There’s no excuse to use anything other than qemu, in my books. But I can admit that it might be a bit obtuse to understand at first. qemu’s greatest strength is also its greatest weakness: it has so many options that it’s hard to know which ones you need just to get started.
qemu is the swiss army knife of virtualisation, much like ffmpeg is the swiss army knife of multimedia (which comes as no surprise, given that both are written by Fabrice Bellard). I run a dozen permanent VMs with qemu, as well as all of the ephemeral VMs used on builds.sr.ht. Why is it better than all of the other options? Well, in short: qemu is fast, portable, better supported by guests, and has more features than Hollywood. There’s nothing other hypervisors can do that qemu can’t, and there’s plenty qemu can that they cannot.
Studying the full breadth of qemu’s featureset is something you can do over time. For now, let’s break down a simple Linux guest installation. We’ll start by downloading some install media (how about Alpine Linux, I like Alpine Linux) and preparing a virtual hard drive.
curl -O https://nl.alpinelinux.org/alpine/v3.8/releases/x86_64/alpine-standard-3.8.0-x86_64.iso
qemu-img create -f qcow2 alpine.qcow2 16G
This makes a 16G virtual hard disk in a file named alpine.qcow2, the qcow2 format being a format which appears to be 16G to the guest (VM), but only actually writes to the host any sectors which were written to by the guest in practice. You can also expose this as a block device on your local system (or a remote system!) with qemu-nbd if you need to. Now let’s boot up a VM using our install media and virtual hard disk:
qemu-system-x86_64 \
-enable-kvm \
-m 2048 \
-nic user,model=virtio \
-drive file=alpine.qcow2,media=disk,if=virtio \
-cdrom alpine-standard-3.8.0-x86_64.iso \
-sdl
This is a lot to take in. Let’s break it down:
-enable-kvm: This enables use of the KVM (kernel virtual machine) subsystem to use hardware accelerated virtualisation on Linux hosts.
-m 2048: This specifies 2048M (2G) of RAM to provide to the guest.
-nic user,model=virtio: Adds a virtual network interface
controller, using a virtual LAN emulated by qemu. This is the most
straightforward way to get internet in a guest, but there are other options (for
example, you will probably want to use -nic tap
if you want the guest to do
networking directly on the host NIC). model=virtio
specifies a special
virtio
NIC model, which is used by the virtio kernel module in the guest to
provide faster networking.
-drive file=alpine.qcow2,media=disk,if=virtio: This attaches our virtual
disk to the guest. It’ll show up as /dev/vda
. We specify if=virtio
for the
same reason we did for -nic
: it’s the fastest interface, but requires special
guest support from the Linux virtio kernel module.
-cdrom alpine-standard-3.8.0-x86_64.iso connects a virtual CD drive to the guest and loads our install media into it.
-sdl finally specifies the graphical configuration. We’re using the SDL backend, which is the simplest usable graphical backend. It attaches a display to the guest and shows it in an SDL window on the host.
When you run this command, the SDL window will appear and Alpine will boot! You
can complete the Alpine installation normally, using setup-alpine
to install
it to the attached disk. When you shut down Alpine, run qemu again without
-cdrom
to start the VM.
That covers enough to get you off of VirtualBox or whatever other bad hypervisor you’re using. What else is possible with qemu? Here’s a short list of common stuff you can look into:
- Running pretty much any guest operating system
- Software emulation of non-native architectures like ARM, PPC, RISC-V
- Using
-spice
instead of-sdl
to enable remote access to the display/keyboard/mouse - Read-only disk images with guest writes stored in RAM (
snapshot=on
) - Non-graphical boot with
-nographic
andconsole=ttyS0
configured in your kernel command line - Giving a genuine graphics card to your guest with KVM passthrough for high performance gaming, OpenCL, etc
- Using virt-manager or Boxes if you want a GUI to hold your hand
- And much more…
There’s really no excuse to be using any other hypervisor1. They’re all dogshit compared to qemu.
-
Especially VirtualBox. If you use VirtualBox after reading this article you make poor life choices and are an embarrassment to us all. ↩︎
Have a comment on one of my posts? Start a discussion in my public inbox by sending an email to ~sircmpwn/public-inbox@lists.sr.ht [mailing list etiquette]