virtualbox.org

This is kind of historic research in virtualization area, which I do from time to time.

Hi All !

The Open-Source world has created much more virtualizers than I ever imagined possible And more so than the proprietary world !

Proprietary world virtualizers:
-VMware Workstation
-MS VirtualPC
-Parallels Workstation
-Serenity Virtual Station (Product discontinued; They even recommend their customers to migrate to VirtualBox ! yay !)
-Win4Lin

All of the OSS virtualization products are partially based on at least one of the two grandfathers:
-Qemu
-Bochs

The more famous ones are:
-Xen
-KVM
...and
-VirtualBox ofcourse... !

I would like to hear your point of view about the *less* known ones:
-plex86 (by Kevin Lawton)
-User-Mode Linux (UML)
-lguest
-KQemu
-QVM86

1. Has anyone any experience with those ?

2. More direct question: How are the less known ones compare to each other ?

3. What were plex86 capabilities?
Some info here:
A. http://www.linux.ie/old-list/25407.html
B. http://www-ti.informatik.uni-tuebingen. ... Plex86.pdf
C. http://linux.slashdot.org/article.pl?si ... 12/2059250
D. http://www.usenix.org/publications/libr ... lawton.pdf

update: ahh, OK, it seems plex86 had 2 generations:
Plex86 1st generation (aka FreeMware) was released in year 2000. It was a full virtualizer (like VMware), capable of running Windows 95 and any other guest. This first generation is dependent on the host kernel driver. Linux host only. See paper "D".
Plex86 2nd generation (aka Light-weight Plex86) was released in year 2002. It was stripped down to become a lightweight para-virtualizer,capable of running only modified versions of Linux guests, similar to LGuest or Xen (PV mode).

Those two were basically different products, yet made by the same man, and shared the same project name. (which is confusing for small history learners as myself)
Is there any evidence of using kernelspace code on the host ? Any evidence of running on non-Linux hosts ?

Back in year 2000, Kevin Lawton dreamed of many now completed features !
Such as: para-virtualization, python scripting, snapshots, differencing hard disks, AMD Hammer support (x64), Cloud computing, etc... (see paper "D")
Great Vision !

-Technologov

Sorry for hijacking your post for an indirectly related question. I would like to know how does KVM compare with Virtualbox. In terms of performance and/or usability. Thanks in advance for any input.

KVM is a kernel module accelerator, similar to "vboxdrv". The full system is called Qemu/KVM, similar to "VirtualBox".
Qemu/KVM has fast IO. Disk and network - but slower graphics, which makes it less optimal for desktop users. KVM has very good memory management (swap+KSM), and it supports bigger amounts of RAM per VM. VirtualBox closes part of this gap with v3.1 release, with VirtIO support. See:
http://forums.virtualbox.org/viewtopic.php?f=15&t=24477

Also KVM is limilted to Linux host, which is not acceptable by most users. (whom need Windows host)
Another big problem is that KVM requires hardware VT to work, otherwise you drop to pure Qemu, which is 10x slower.
As for usability - Qemu/KVM lacks GUI, and third-party OSS GUIs such as AQEMU and virt-manager are not good either, so VirtualBox is more usable.

VirtualBox is the world's first and the only Open-Source User-Friendly Virtualizer !

Thank you for your very informative answer. Strange that Redhat and Ubuntu are pushing for KVM while neglecting the friendlessness aspect. I guess Virtualbox will enter the Server market after it is well stabilized in the Workstation market, would VB have enough technical arguments against KVM?

AntiMatter: I do not know reasons for their choice.

Let's discuss VirtualBox history a bit.
I wrote a VirtualBox article 2-and-a-half years ago: (This covers VBox 1.3 on openSUSE 10.2 and 10.3 hosts)
http://forgeftp.novell.com/lfl/.html/virtualbox.html

And it looks great !
What has improved since then?

In my article I touch several hard points, that is features that were very difficult to use in VBox 1.3 are now a matter of clicks in VBox 3.x. This is great achievement.

The following features were considered "advanced" and needed knowledge back then in v1.3:
-Networking via Host Bridging
-Seamless Mode Virtualization
-Backup, Restore and Porting of VMs Between Physical Machines (now it became import/export via OVF)

Some topics, that are still hard: (therefore still needs documentation)
-How-To USB + openSUSE 10.2

I have re-read this article from 2007, and was positively surprised to see one name: "Pablo Sanchez", whom helped me back then to cover some topics. He is still here with us on the mailing-list. Big thanks for such long-term dedication.

I learn history, because I would like to know where-from we came, and where we go. We go into the possible future:
http://forums.virtualbox.org/viewtopic.php?f=9&t=24627

-Technologov

With Bochs you can emulate a defect PC in real time

I checked Bochs some years ago. But its not usable in a way like VirtualBox or VMWARE. The processor is emulated. This is bad for performance, but good for checking features for other processors. You can run Software and a OS not made for your hardware.

In that time there was a first approach to integrate plex86 into Bochs for more performance. It was like kqemu for qemu. I checked this, but the performance was not better. And there was only a patch, no further development. So Bochs died for my use cases.

I bought VMWARE 1 and 2. There was a hobbyist program, you could order it for about 50 $. It was a great product in a time no one was speaking about virtualization. But with VMWARE 3 the program stopped. I had to buy a license about 200 $, but that was to much for software to play with. For me virtualization is a playground like a model train, in germany we called it "märklin"

Parallel there was a company Netraverse with the product "Win4Lin". This was a solution like "WinOS3.1", Windows 3.11 in OS/2. Its started a virtual DOS integrated in the linux kernel and on that a windows 98 was running. But after some time the company go away, there were several following companies. The product died, they didnt develop the patches for newer linux kernels. That was bad, because it was a good product, cheap for private use (about 30-50 $), much performance.

At that time QEMU was growing to a big piece of software. And so i used Qemu, later with KQEMU.

The Win4Lin Company "developed" a new product "Win4Lin Pro". As i know Qemu, i must lough out loud. This was only Qemu for much money, nothing else. The performance was bad because there was no virtualization at the beginning, only emulation.

If you read the homepage from the company, you think they were the explorers from virtualization like Bill Gates founded the internet

Today the situation is perfect for private users. With VirtualBox and VMWARE Player there are great products for no money.

Some time i chcked server virtualization, too. I set up a Xen Server, but its much complicated. You read many articles about giving PCI Devices to Guest. But after playing around and reading much you seee: This is only for virtualized guests with special kernels.

And i checked KVM, too. It has much performance.

But for all OSS Products there are no special tools for good integration to the desktop like VBOX Guest Additions or VMWARE Tools.

So at the moment i used these two solutions and nothing else.

AntiMatter wrote:Strange that Redhat and Ubuntu are pushing for KVM while neglecting the friendlessness aspect. I guess Virtualbox will enter the Server market after it is well stabilized in the Workstation market, would VB have enough technical arguments against KVM?

I don't think the main reason is technical argument, but license argument. VirtualBox is not open-source project, but it is actually open-core software. The main difference between open-source projects and open-core projects is that open-source projects have all of the features open-sourced, but open-core has some core features open-sourced but additional features must be paid - in VirtualBox case it has to be paid for business - read VirtualBox PUEL license. For example if you would like to have remote access for business to single virtual computer for more then one user, you must pay a license fee to Sun.

I am not saying that open-core projects are bad, if company likes to compete to closed-source projects it must get enough money to pay for development process. This is just a business model. There could be another business model to have all of the programs code released under open-source license and so just have a money from support agreements. In this case I think there would be magnitude of less money coming into project because additional features are for companies, that most probably have enough employees to manage/administer virtual machines by they own, so not needing support agreement at all. Imagine Wikipedia project it uses Red Had Linux and must pay a fee to get software upgrades. But they have employees that know Linux in depth, so they decided to migrate to Ubuntu Server which has no fee for software and upgrades - there's business model is to sell support agreements, but in this case Canonical (business company behind Ubuntu project) will get zero money.

The reason why Red Hat supports KVM is logical. Red Had purchased Qumranet company that was behind KVM project in the first place. So Red Hat "owns" KVM. Why would it support a competitors (Sun's) software?

There is also one other reason I think VirtualBox is mainly targeted to desktop computer users and KVM to server users. Both products are available in Ubuntu repository and can be installed without any big hassle. VirtualBox in Ubuntu repository is actually VirtualBox OSE edition, currently in Karmic is VirtualBox 3.0.8. Until VirtualBox will get marketed as server product I don't think many companies will look at it as server product.

VMWare licenses are not cheap, they still have the majority of the server market. I still believe that a metal hypervisor type 1 is more complicate technically than a host based virtualization (hypervisor type2). If I were the Virtualbox owner and I have a server version right now, I will give it for free, just for the sake of grabbing a market share. Admittedly I don't know the implication of licensing scheme with regards of the success of a product or not, but I am pretty certain that if Virtualbox "Server" is not affordable, it will probably have no chance. Take Virtualbox workstation as an example, the only reason I spent time to get acquainted with it is because it is free. My co workers know about Virtualbox because I mention it to them (we use VMWare). At home, my relatives and friends know of Virtualbox b/c they see me using it. All this "radiation" is worth much more than the cost of a license.

Anyway, I think Virtualbox will have some offering in the server market. They better hurry because the server market is already pretty crowded. And once people start investing in training for a virtualization server product, they won't change their mind easily.

AntiMatter wrote:If I were the Virtualbox owner and I have a server version right now, I will give it for free, just for the sake of grabbing a market share.

VMware is doing this very aggressive. I hope VirtualBox will get more open too.

AntiMatter wrote:They better hurry because the server market is already pretty crowded. And once people start investing in training for a virtualization server product, they won't change their mind easily.

Virtual products are already crowded inside Sun and Oracle. I don't believe all of them will be offered in the future.