Hi,

I'm thinking about the possibility of buying a 4+ core computer, with 8+ GB of RAM, so that two (or more) people could use the same machine at the same time, instead of using separate computers as right now. I could use some kind of KVM switch to be able to connect two or more keyboards, mouses and monitors.

But, what about the OSes? I want Slackware as the host system and the next Ubuntu 14.04 as the guest, and I'd like to avoid full virtualization solutions. Vmware and VirtualBox are out of question.

For the 'light' virtualization solutions, I've read that the kernel of the host should match the kernel of the guest, what seems very difficult to attain when considering Slackware and Ubuntu.

As there are some experts on this virtualization area at this forum, and there are a bunch of technologies available (KVM, OpenVZ, LXC, heroku, docker, coreos, lmctfy, zerovm) I'd like to know if such a solution is somehow possible.

Let me stop you right there, because that's not going to work. Short of patching the kernel or using custom-made drivers, there's (currently) no (good) way to tie an input device and a monitor to a session.

You could set up a server and have the users connect over the network, of course. Then several users could use the same system simultaneously, but they'd need some sort of local client, like a PC or a thin client.
Why would you want to avoid the fastest and most secure virtualization solutions on the market? With the introduction of CPU extensions for hardware-accelerated virtualization, performance is better than ever. You can further improve on that by using paravirtualized device drivers for disk and network access.

That's the whole deal with containers (which is what you're talking about); they run under the same kernel, one that supports namespaces to keep containerized userspace processes segregated.

Ubuntu uses a patched kernel, but I don't know to which extent it actually requires those patches in order to run. Some of the patches relate to hardware drivers, and in a containerized environment, drivers are a non-issue.
I've been configuring, installing and using hypervisors and VM systems on the x86 platform since the early VMware days back in the late 90s (and had to listen to the SunOS/Solaris guys telling us just how far behind we were), but I'm certainly no expert.

I can recommend KVM, as it's a fast, robust and fully-featured hypervisor. Add libvirt, and you've got a system that's quite easy to manage. On the container side I've only ever really used LXC, but it certainly seems to work well enough and doesn't require kernel patches (which turned out to be a problem the last time I made a serious attempt to evaluate OpenVZ, but admittedly that was a few years ago).

Humm, I didn't try this device but it seems to enable multiuser access to the same workstation. Since Linux is also multiuser, I was hoping to be able to have more than one person connected to the computer, taking advantage of different sessions (also X sessions). One of them could be connected to the guest operating system. Or am I wrong about this kind of products?

Some time ago, I installed VirtualBox in Slackware64 and didn't like it at all, I mean the fact that it didn't have a 64 version and also the deinstallation part. I use VMPlayer in my current computer, but would like to use an open-source setup only for this system. So I guess, I have qemu as the only solution...


Yes I know it. And besides installing the systems, keeping them updated would probably be a nightmare.


I'll take a better look at OpenVZ.

Well, I'd disagree. Using KVM you can passthrough any PCI devices, including USB ports for keyboard and mice and VGA cards to support another monitor. No new hardware required, except a chair..

I am currently running Slackware 14.1 as my host and Win 7 as a VM guest; each has it's own monitor and graphics card, and its own mouse. You can also use Synergy to share a mouse and keyboard if you prefer. The more traditional VM approach is to use VNC, but separate hardware (mouse keyboard, screen) with a multiseat approach is quite feasible without any physical switches.

Although I have a Windows guest, There's absolutely no reason why you can't do the same with a linux or BSD guest; in fact it should be easier considering how persnickety Windows can be about hardware drivers.

Yes, but that's not going to work if the OP insists on not using a hypervisor, is it?

I tried using the "passthrough" features of KVM about three years ago to make a USB plotter available to a Windows guest, but back then it was horribly unstable. The printer would work for a while, typically a few days to a week at the most, but then communication would be lost and only a reboot of the hypervisor could make it reappear.

I also tried using PCI passthrough to access a Matrox video card, but gave up after repeated kernel panics on the hypervisor. Glad to hear it's working properly now.

True enough, OP needs to use virtualization or the whole thing doesn't make sense. Closest thing I can think of otherwise is coLinux under Windows, where the kernel runs as a service...

KVM has certainly matured tremendously in the last year; I only used VBox and VMWare a few years ago. Passthrough seems to be pretty stable for me at least, but there's no question it's still kind of a bleeding edge thing, as opposed to a click on an icon in Ubuntu thing. (Not that's there's anything wrong with that.)