undefined | Better HN

0 pointsasciimov2y ago0 comments

Thanks for the detailed write up.

I don't have full notes for my last adventure into QEMU, but you are correct that it was mostly an issue with libvirt and getting graphics to render correctly.

> Well, given how long this post is, it certainly doesn't do anything to disprove the notion that Qemu is less accessible and intuitive to use for ordinary users than Virtualbox, that's for sure. And to be sure, I 100% agree and empathize with the fact that it is not, because it isn't.

That's really the rub, choosing to be an ordinary user. The older I get the less I choose to tinker with things that I'm not going to use very often. I like it when software just sets itself up and lets me get to work.

0 comments

jchw2y ago

This is true. If someone says "I just want something that works, and Virtualbox works" it doesn't really raise any eyebrows. It's really the tone of "I am unable to make QEMU/KVM work, since I am not a rocket scientist" that implies, maybe they would use QEMU, if only it wasn't horrifically difficult to get working. I think if your experience with QEMU is this bad, you may have just gotten a bit unlucky and gotten off into the weeds on a detour you probably did not need.

That said, are there reasons why someone who just wants things to work and doesn't want to tinker ever would choose QEMU over Virtualbox? Definitely. Here's a good one: the Virtualbox kernel modules were notoriously problematic. Back in the day, loading the vboxdrv module would add TAINT_CRAP in your running kernel[1], because the kernel developers were tired of dealing with bug reports that are the fault of the vboxdrv. Presumably it's better nowadays, but out-of-tree modules are generally a source of headache. Another good one is features: Virtualbox has a lot of great features for desktop end users, but QEMU can do a whole lot more; it supports absolutely tons of architectures and system options. You get access to a wide range of paravirtualization devices through Virtio drivers, including VirtioFS, a dramatically superior solution for mounting directories into VMs versus Virtualbox's Shared Folders feature, in my opinion. It's also possible to do a lot more advanced things, like setting up a discrete GPU passthrough and Looking Glass, or passing a portion of your Intel iGPU using GVT-g.[2]

Of course, some day Virtualbox may support a KVM backend, just as virtualization tools are starting to support unified hypervisor backends on Windows and macOS. There's even an existing patch, though I do not know what the status is and whether or not it could ever be merged upstream.[3] So maybe in the future, choosing between Virtualbox or KVM will become unnecessary.

For now though, it's definitely not clear what you should recommend to someone IMO. (For this particular use case, I don't even recommend a VM at all; I think for Windows 95, 86Box is a better solution.)

[1]: https://lkml.org/lkml/2011/10/6/317

[2]: https://wiki.archlinux.org/title/Intel_GVT-g

[3]: https://github.com/cyberus-technology/virtualbox-kvm, https://news.ycombinator.com/item?id=39300317

gonzodaruler2y ago

1 point by gonzodaruler 0 minutes ago | next | edit | delete [–]

> Of course, some day Virtualbox may support a KVM backend, just as virtualization tools are starting to support unified hypervisor backends on Windows and macOS. There's even an existing patch, though I do not know what the status is and whether or not it could ever be merged upstream.[3]

This is the official patch set for the VirtualBox KVM backend. Our plan is to support this for the near future. There are even packages for ArchLinux (AUR), gentoo and NixOS already if you want to test this.

j / k navigate · click thread line to collapse