Weekly Edition
Return to the Press page
| |||||||||||||
|
| |||||||||||||
Introduction to the Xen Virtual Machine (Linux Journal)
Rami Rosen
introduces Xen in a Linux Journal article.
"The Xen VMM (virtual machine monitor) is an open-source project that is being developed in the computer laboratory of the University of Cambridge, UK. It enables us to create many virtual machines, each of which runs an instance of an operating system.
These guest operating systems can be a patched Linux kernel, version 2.4 or 2.6, or a patched NetBSD/FreeBSD kernel. User applications can run on guest OSes as they are, without any change in code. Sun also is working on a Solaris-on-Xen port."
(Log in to post comments)
9 Posted Sep 1, 2005 20:37 UTC (Thu) by ncm (subscriber, #165) [Link] I like that you can run vanilla Plan 9 on it, too.
To me this means that the Plan 9 project can forget about fooling with device drivers, and concentrate on improving the kernel/user-level interface. C89+Plan9 could become a far, far better portable virtual machine than JVM or CLR, given just a few tweaks (e.g. guaranteed tail-recursion-optimization, and official definitions for commonly-depended upon undefined behaviors).
Don't forget Inferno and 9P Posted Sep 1, 2005 21:26 UTC (Thu) by Zarathustra (guest, #26443) [Link] Talking about portable virtual machines and Plan 9, don't forget Inferno and Limbo, which have been recently relicensed under GPL/MIT.Some people are also pushing 9P as the standard protocol for exporting devices in hypervisor systems, that would provide many advantages, like network transparency and OS/arch agnostic communication, and will work great with projects like v9fs(which hopefully will go into the main Linux kernel soon). Plan 9 has demonstrated that a file system interface is a very powerful and flexible representation of resources, it's time for other systems to start picking up those ideas.
Don't forget Inferno and 9P Posted Sep 2, 2005 0:56 UTC (Fri) by ncm (subscriber, #165) [Link] ... don't forget Inferno and LimboI already have, thank you. (Java itself was bad enough.)
Don't forget Inferno and 9P Posted Sep 2, 2005 7:45 UTC (Fri) by Zarathustra (guest, #26443) [Link] Are you implying that there is anything wrong with Inferno and Limbo? Limbo is probably the best general purpose language ever created, and very much in the spirit of C and Unix. Inferno is a great VM which is extremely portable and lightweight, and can run both native on hardware and hosted on almost any other OS.Both designed and developed by the same team that created Unix and Plan 9. Limbo, together with Erlang has the only decent concurrency model of any non-academic language, I still can't understand why everyone else ignores the wonderful CSP model. Ah, and then there is all the distributed computing stuff which was based on Plan 9 and is light years ahead of anything else.
Don't forget Inferno and 9P Posted Sep 5, 2005 18:40 UTC (Mon) by sbergman27 (subscriber, #10767) [Link] Sorry, Zarathustra. And I know that Plan 9 and Inferno are near and dear to your heart; You champion them at every opportunity. But they, and their associated components, have nearly zero mindshare outside of the few and the faithful.
I can't speak for other Linux fans, but I feel I am in a small enough minority as it is.
Plan 9 might be the best designed OS ever created. And it doesn't make a whit of difference in the real world.
Introduction to the Xen Virtual Machine (Linux Journal) Posted Sep 2, 2005 1:00 UTC (Fri) by mmarq (guest, #2332) [Link] Let me consider here the main use of a virtual machine in Linux essencialy only as a mean to run MS Windows OSes and apllications.
" These guest operating systems can be a patched Linux kernel, version 2.4 or 2.6, or a patched NetBSD/FreeBSD kernel. User applications can run on guest OSes as they are, without any change in code... "
- All of them are GPL compatible,... yet they dont run the Monopoly applications and are not consequently adequated to 'embrace & extend'...
- ReactOS:
- Is two licensed under the GPL
- Is not a modification of WindowsXP or a clone, but a total new implementation relying heavly on WINE, so no licence or patent issues should arrise.
- Considering XEN Save Hardware Interface with sharing and isolation:
- ReactOS provides many of the features that actual implementations of WINE lack, including a better Registry than Registry and others, so fundamental for having much more Win32 applications running on it, than possible with WINE only.
- And analogue to ReacOS heavy relyance on WINE as an OS, i belive it could also be made (in a XEN port) to 'link' not only with WINE but with a running Linux shell, filesystems, network, *whitout that annoying inicial shell window* so caraterist of having an OS on top of other OS, and with the same 'look & fell' allover, if a proper win32-gtk-qt engine for Linux themes would be developed and applyed. In a word TRANSPARENT Win32 apps on Linux.
- Thought the ReactOS developers goal is to build a complete independent OS, i belive they would be more than thrilled to have extra eyes on their code. ReactOS could also 'integrate' Xen... but better, that is, they instead of clonning the windows driver model filesystems and network, positioning themselfs for law suits, could use the Linux ones, having an OS that TRANSPARENTLY runs win32 and linux apps if the proper KDE(dont have a link but belive a port of KDE to Windows is being attempted,... great!) or Gnome or other shell is applyed. No wars, no tricks, no unpoliteness... simple cooperation would be very easy to achieve, i belive, because of so many advantages to everybody.
" Another known project is qemu. The disadvantage of hardware emulators is their performance. "
- No 'qemu' also can be a full virtualizer. It would be interesting to see comparative benchmarks. A simple ang good description here: http://flaviostechnotalk.com/wordpress/index.php/2005/08/...
" As was said in the beginning, currently Xen patches the kernel. But, future processors will support virtualization so that the kernel can run on it unpatched "
- Qemu already interfaces an unpatched linux kernel or MS OS, which can be advantageous for simplifying installations. But When/If Xen delivers as one default domain an integrated port of ReactOS, representing the possibility for Linux to integrate the large majority of the IT deployment out there, and with the added comfort of a Safe Hardware Interface, why worry about that requirement when distros are already delivering it everywhere.
" At boot time, Xen is loaded into memory in ring 0... Currently, domain 0 can be a patched 2.4 or 2.6 Linux kernel. According to the Xen developer mailing list, however, it seems that in the future, domain 0 will support only a 2.6 kernel patch "
- With the proper FULL integration of Xen in the official kernel tree, i belive there would be no patching worries, because every Linux kernel would be Xen ready. Also full integration will open doors to merging RTAI/Fusion, which is a layer on top of Adeos nanokernel in ring 0, nothing imposing that it wouldn't be in future on top of some Linux naokernel http://lwn.net/Articles/106010/.
-RTAI/Fusion runs also in a hypervisor mode similar to Xen, so is kind'a logic that an integration between the two under the benevolent dictatorship of Linux could be possible and achived without clashes... not easy but without clashes.
The end will be a OS capable of full virtualization, deterministic real-time, with a superior class, isolated, safe and hardened driver model, and capable of transparently running most win32 apps... the nirvana, no ?!
Introduction to the Xen Virtual Machine (Linux Journal) Posted Sep 2, 2005 1:43 UTC (Fri) by mmarq (guest, #2332) [Link] Sorry to comment myself...
" RTAI/Fusion runs also in a hypervisor mode similar to Xen, so is kind'a logic that an integration between the two under the benevolent dictatorship of Linux could be possible and achived without clashes "
Tought far from being an expert, i belive that is clear that the point i've being trying to make here is technical in nature and not political.
Xen prescribes an isolated, hardened(crash free) environment por drivers. RTAI/Fusion also prescribes some how a separeted layer for deterministic real time and no, being my opinion that would be tremendously advantageous to have all *hardware device drivers* (not filesystems) in a deterministic real-time mode, with different time requirement dependending on the systems ,sub-systems and applications, but hard real-time none the less.
With hotplugging going everywhere, with asynchronousity also going everywhere, and with the widespread of everytype of "plug & play" devices and with the use of more than 1 processor protected memory addressing rings, its each time more evident a political artifiality for keeping all drivers close tight in the tree.
Better with a proven crash free enviroment for drivers, it will fall and smash's itself in the ground the principal objection for out of tree drivers.
Introduction to the Xen Virtual Machine (Linux Journal) Posted Sep 2, 2005 14:09 UTC (Fri) by proski (subscriber, #104) [Link] I believe new processors (see Vanderpool and Pacifica) will make it much easier to run guest OSes without modification. Reactos is an important project, so let's not distract its developers from their current work. Maintaining a separate kernel would be an extra burden for them.
Introduction to the Xen Virtual Machine (Linux Journal) Posted Sep 2, 2005 19:13 UTC (Fri) by mmarq (guest, #2332) [Link] " ... Maintaining a separate kernel would be an extra burden for them. "
Sorry, but that is "La palisse"...
|
|
||||||||||||