while there's no practical case for doing this, perhaps beyond 2 levels deep, sometimes some people do this just to see if they can or to see how much it slows down, i'm curious who has tried this and how deep they got with which virtualization technology on what physical hardware host (or cloud provider as the case may be).

why? embedding VMs into VMs is possible, depends only on the available resources (if you have enough resources to start a new VM). But as you told there is no practical case. It is extremely inefficient.
If I remember well we created an openstack environment inside a VM, probably that was 2 or 3 levels deep (altogether). Also started to test some software inside, but unfortunately I can't remember the details. It was definitely not a performance test. Usually we want to avoid that situation.

> why?

why not.

if you are into systems or system setup or such, you may want to just play around. i did back in my mainframe days when i had an IBM 370/158 (8 MB memory) all to myself. i managed to run IBM's VM/CP (runs directly on hardware or as a guest OS in a VM) 6 levels deep before i ran out of spare disk space. it was a lot slower by this point. and it bogged down the host system rather hard, too, which is why i didn't try this on our primary machines.

i am also curious how well VirtualBox performs like this compared to QEMU/KVM with various Linux setups.

it's all for fun. those who have never done fun things on their own computers need not bother trying to answer in this thread.

Most modern VM's have pretty good resource management. You just have to have almost unlimited resources to run almost unlimited nested VM's. Each VM has a bit of overheard and then the OS.

Nothing can get past the need for resources.

I run embedded VMs at work daily. Doing it as we speak. My attack box (jump box) is a Windows box which is a VM hosted on VMware (huge enterprise host somewhere). On that jump box, I run VirtualBox which hosts my Kali instance. Works perfectly That's as deep as I go though - no point in any more for me anyway.

You limit the resources of every VM and you need to divide these resources among the embedded VMs. That may mean you will be low on resources in the 3rd level, because there will be no [enough] resource to manage.
From the other hand every VM has its own [virtual] memory manager and that means on the 4th level [of nested VMs] you will have 4 (or 5) memory managers working "together" to try to serve its processes - without knowing anything about the other VMs. That will make the full system unstable, unreliable and most probably useless.

Embedding VMs only one by one into each other is just wasting resources (like in post #5), but can work without problems - if you have enough resources to waste.

so you are saying that i can run as many levels of VMs as i have resources for?

i have about 16GB of RAM and 1TB of storage to play with. how many levels do you think someone could achieve with that and their choice of distro and VM software (same at each level except the first host, configured best for this effort)? a VM that need a zillion libraries would be one that i would avoid.

i already "play" with resources like trying to run the largest buffers i can in a process to see how slow everything else comes to be. i do other kinds of resource consuming "playing around" like that when not running anything serious on here (as i am right now, running 3 concurrent copies of ffmpeg for the past 2 weeks reformatting over a hundred MP4 videos into WEBM/VP9 format to save space).

each level of VM will need to be configured with enough resources to run all the levels it hosts. each level would be smaller and smaller until the last can barely run helloworld.

i used to do embedded systems at a previous job. but there were several eval boards to run on direct hardware. i don't do embedded, anymore.

are you running an embedded distro as the guest OS?

Yes. The host is on VMware and the guest is a VM on VirtualBox.

Each level consumes resources. You can keep installing what ever you want till it stops.

1 members found this post helpful.

In my case, most of our servers are virtualized on VMWare. I and each of my testers has their own server jump box that we attack hosts from. On each of those virtualized hosts, we run Kali Linux in VirtualBox. Performance is near bare metal, as far as I am concerned. I have no idea what the physical VMWare hosting server is though, but I believe it is quite beefy.

I thought about this question a long time ago. I can give it a try.

Research problem:

Virtual machine nesting ability test:

1. Is infinite nesting possible?

2. Actually feasible nesting level.

Conjecture: VMWare supports unlimited nesting.
Because VMWare supports virtualized VT and VTx, it is speculated that in theory, it can be nested infinitely. If the virtual machine's ability to "virtualize" does not decrease as the number of nesting levels increases, this assumption is true.

Experiment procedure:

The first layer is no problem, VMWare is successfully installed in the virtual machine.

Then when running VM-XP2, the splash screen was loaded, and then the VM crashed. There is no documentation for this magical error.

But the Linux virtual machine can run normally, and it runs in the VT-x/EPT virtualization mode

Therefore, it is certain that multi-level nesting is actually feasible. In fact, it is of course possible to use pure software to simulate hardware operations, such as Bochs. But the performance will drop by several orders of magnitude. The performance degradation of VM-XP is too severe. I think changing to VMWare, Hyper-V, VirtualBox, Xen or even Bochs with Linux and Windows may be better. Forget it, I can continue to do it anyway... I have lost the interest in studying how many levels can be nested, it doesn't make much sense anyway.

You can't get infinite unless you have God's computer.

You will quickly find this task to fail. It basically useless to attempt it unless you just want to see. There is no practical use for it.