Talk:Xen

From Wikipedia, the free encyclopedia

Contents

[edit] Plan 9 and virtualisation

it seems that unix cope it's limitation with virtual machines...

plan 9(open-source and fsf aprooved) achive the same thing(jailying apps inside some resource) in a different way... see here mabe we can talk about theses limitations and tell that virtualisation is the most used solution but that others exist...

I think that discussion belongs on a more general 'virtualization technologies' page.--LukeCrawford

[edit] Dom0 and Domu

http://wiki.xensource.com/xenwiki/OSCompatibility

what are they

dom0 as said in this talk page is the host os...and domU should be the guest os.

[edit] Virtual wars?

Hmm. I wonder whether it'd be relevant or neutral to mention that Xen seems to have won the FSF wars with regard to virtualization? ex: Xen gets some endorsements and support, and a startup- [1]. -- Maru Dubshinki 08:31 PM Thursday, 03 March 2005

I think that the XenSource startup company might be worth a mention, since it is basically the "xen company". As for being the winning any wars, I think it has a good chance of being the definitive open source virtualisation solution. However, I don't know that it has (yet) exceeded User Mode Linux, etc. in terms of deployments. I am also arguably biased about how good Xen is, so I've been sticking to largely technical content so far ;-) --Mark Williamson
I am actually wondering whether thinking of it as 'virtualization wars' is a red herring/moot point, what with the imminence this (~2005) year of AMD's and Intel's Pacifica and Vanderpool, respectively, techonologies. Actually, come to think of it, didn't AMD also contribute to Xen as well as Intel? -- Maru Dubshinki 12:49 PM Friday, 01 April 2005
Well, there'll still be competition between different virtual machine monitors sitting on top of those technologies. AMD have, I think, said that they will contribute to Xen but I don't believe they've made an actual code drop yet (Intel have already contributed a substantial amount). --Mark Williamson

I agree that there is a war... there is :

  • qemu with the "proprietary-freeware" kernel module
  • specific virtualisations such as uml
  • Xen
  • vmware "proprietary with free "testing" branch and free client"

look atentively at xen benchmark against uml...uml is a 2.4 kernel!!!...and uml patch has been integrated and avaliable by default inside the 2.6 branch since 2.6.??

--- It's not a war so much as a bake-off...

So how does the inclusion of KVM in the Linux kernel, as well as the announcement that Fedora Core 7 will include KVM effect Xen's position? If Linus Torvalds and RedHat are behind KVM, how long will RedHat remain in both camps before they decide to "officially support" one project over the other? --Alex Weeks

[edit] ring 1

I'm sure I read that zen ran the guest operating systems in ring 1. if this is true then a mention should probably be added to this article and to the rings 1 and 2 article. --Plugwash 11:37, 6 Jun 2005 (UTC)

I had thought that that was how it ran with virtualization extensions- Xen (or other hypervisors) would run in Ring 0, and the unmodified OS would run in Ring 1, all happy and unchanged. The modified OS are necessary when you have to run Xen in Ring 1 and force the OS into Ring 2, where an unmodiifed OS is unused to and would muck things up. --maru 17:21, 6 Jun 2005 (UTC)
Actually, Xen normally does run guest operating systems in Ring 1, which is why they need to be modified. When there is hardware support, a new axis is added: you have rings 0-3 and *also* "root mode" and "non-root mode". The hypervisor and other virtualisation software run in rings 0-3 of Root Mode (the hypervisor itself runs in ring 0, root mode) and see the real hardware. The guest runs in rings 0-3 of Non Root Mode, and so is prevented from seeing the real hardware and under complete control of the hypervisor. And yes, all of this information should probably be in the article somewhere ;-) I might get around to adding it later but if anyone else takes a crack at it then I can review it (I work on Xen fulltime). Cheers guys! --Mark Williamson 00:06, 7 Jun 2005 (UTC)
I was wondering how exactly you knew so much... But since I don't seem to have the whole ring thing quite right, I'll let you do it. --maru 00:25, 7 Jun 2005 (UTC)
Heh that's my secret :-) It's been great how many contributions this page has had - for months it had about one sentence on it, almost got deleted. I've tried to add stuff where I can but it'd be nowhere near as complete as it is without everyone else - I seem to remember you making a lot of the additions... Anyhow, I'll add the rings thing soon. But please do keep adding other stuff! --Mark Williamson 00:40, 7 Jun 2005 (UTC)
"The Xen host runs in Ring 0, while the guest runs in Ring 1." - Erm, correct me if I am wrong... the host OS Dom0 does NOT run in ring 0 only the Xen code runs in ring 0 not even the Dom0 runs in ring 0 which is why VMWare and KQEMU do not run on Xen kernels..... I believe it can be cited from the xensource mailing lists NthDegree 16:03, 18 February 2007 (UTC)

[edit] Comparison to UserModeLinux?

How do they stack up, what's the main difference? --Anon.

UML =Linux only, less efficient, but does not need kernel modified (the host kernel might need to be, I'm not sure, but I'm fairly sure the guest kernels can run straight), more tested etc. --Maru 12:42, 24 August 2005 (UTC)
UML certainly does require modified guest kernels although uml is now part of the official kernel tree so its just a compile option now.
The host kernel doesn't need to be modifed for uml but uml can run a lot more efficiantly if it is. --Plugwash 13:02, 24 August 2005 (UTC)
/shrugs. Good thing I don't edit the UML article then. --Maru 13:14, 24 August 2005 (UTC)
We definitely should have UML mentioned in there. Mark Williamson 01:41, 31 August 2005 (UTC)

where did you heard that uml was less eficient... look atentively at xen benchmark against uml...uml is a 2.4 kernel!!!...and uml patch has been integrated and avaliable by default inside the 2.6 branch since 2.6.??...so we need to test them...by the way i use gentoo and linux kernel ported to xen is hard maskered(that means that it has a problem...such as security or other...i should ask them if i can use it safely,because some of their masked package can be used safely if you know the problem...for example a security bug in a game(unreal tournament) used as a game server=>if you use it only as client...you don't have problems)

It's well known to be less efficient, although that is only important if efficiency is your concern. With UML, the guest's instructions are being executed in terms of a user space program which means context switching. You need at least two fully-fledged operating systems. With Xen, you have the Xen OS which is pretty small, plus your guests. -- 82.39.205.158 23:58, 29 July 2006 (UTC)

[edit] Windows support

Will Virtualization Technology enable support for Windows as a HOST OS, not only guest? See under "software platforms". Aditsu 16:05, 9 February 2006 (UTC)

Not for the foreseeable future; the "host" (dom0) OS has to be Xen-aware. Theoretically possible to make what you propose work, but not terribly likely to happen any time soon as far as I can tell. Mark Williamson 12:43, 10 February 2006 (UTC)
Then perhaps the phrase "This may change in the near future however, when mainstream processors will support virtualization instructions" should be removed? Aditsu 09:04, 13 February 2006 (UTC)
there was a non-distribuable port of windows xp...but it can't be distributed because of the windows xp licence...lol
by the way in this kind of paravirtualisation approach the kernel must be ported to xen because there are issues such as the scheduler:a normal kernel expect to have all the resources of the machine and so it expect to have the cpu at any time...and in such vistualised environement the cpu can be acessed by another program running on the host os and so it's unavailiable for the xen kernel...
and so VT and such processor features cope with this(i've not the confirmation of this but it seams logical...)
VT doesn't eliminate problems like scheduling - there are still some slightly weird behaviours in terms of dealing with multiplexing OS kernels that don't expect to be pre-empted. The main benefit of VT, however, is that it enables you to emulate instructions which previously couldn't be efficiently emulated, and it has some miscellaneous features for accelerating virtualisation in general. The end result is that it's simpler to support unmodified guests than it was before VT extensions were available, and it should be possible to support them with higher overall performance. Mark Williamson 17:39, 7 March 2006 (UTC)
Unmodified unprivileged guests will run fine under Xen/VT, it's just that the *host* needs to be Xen-aware. i.e. you could run Windows in a guest, with a Linux, or NetBSD dom0 (for instance). So the phrase about stuff changing in the near future should be fine. In fact, Xen can run unmodified guests on Intel VT right now, with the next point release including support for AMD Pacfica also. Mark Williamson 17:39, 7 March 2006 (UTC)
It seems that Micro$oft has seen the light and is now going to offer Xen-host capabilities in Longhorn Server [2]. Of course, this is (IMO) an attempt to increase the market for Windows Server, which would otherwise tip even further in favor of Unix/Linux. Note that Longhorn won't be able to run as a Xen guest, as far as I can tell, since the announcement refers to Linux as the "guest". Vindictive Warrior 22:21, 20 July 2006 (UTC)

[edit] Link to xen-forums

There are too many links in the External Links section. To start with, I'm removing the xen-forums link as it is not very popular (there are only a handful of posts). Anthony Liguori --70.116.9.243 06:21, 31 May 2006 (UTC)

[edit] Popularity contest

With an annual VMM product revenue in excess of US$600 million (as of Q2 2006), VMware is by far the most popular VMM for Windows virtualization.

??? Sales =! Popularity, especially when people don't pay for most of the products listed. Bruce 04:15, 18 August 2006 (UTC)

[edit] The comparison section is unnecessary

1. No actual comparisons are being made. This section only contains 1 or 2 sentence factoids about other VMM. For reference, a comparison is defined as:

  1. The act of comparing or the process of being compared.
  2. A statement or estimate of similarities and differences.

Again, neither of these actions occur in the "comparison" section; what this section does is summarize other VMM's that are on the market.

2. Partly because of the lack of comparisons to Xen, this section does little to further the reader's understanding of Xen (which is the focus of the article). This section needs to be removed from the Xen article and given its own article-space, perhaps under the heading of "VMM synopsis" or some other appropriate title.

For reference, I direct you to the article on OpenVZ. This article has a brief paragraph comparing it to 2 other VMM's, and it makes real comparisons.

[edit] This page needs serious work

A good portion of the info on this page is factually incorrect. I've gone through and marked the most objectional paragraphs as unreferenced but in relately, as a Xen developer, I know the information is wrong. Among the biggest problems are the reference to Xen supporting all x86 Operating Systems (that's based on a quote from Simon Crosby about Xen supporting all *major* x86 Operating Systems--well, he's wrong but leaving out that major qualifier makes a big difference). The information about Microsoft's support for Xen is also wrong. Viridian is the codename for their hypervisor--not some sort of Xen implementation. Microsoft merely licensed the Viridian guest API to XenSource. XenSource is building a shim to allow Xen guests to run under Viridian. Finally, the performance numbers are, at best, uncited. In reality, software bridging alone will cause a 10% performance impact on any sort of network-heavy workload. The true numbers are certainly much higher. Xen is still considerably faster than something like VMware...

I'm going to sit on this for a little bit with the hope that someone proves me wrong and references all of this. Otherwise, I think it's appropriate to rewrite most of the content of this page. -- AnthonyLiguori --70.112.17.156 15:06, 19 September 2006 (UTC)

I've started cleaning up this page. Please be care when adding new content. Wikipedia has policies on external linking that should be consulted before adding. I also removed the comparision section as it's duplicated elsewhere in wikipedia. I've removed any reference to Microsoft shipping a Xen port. They aren't. I've clarified the relationship between XenSource and Microsoft. I've also removed references to an, at worse, 8% performance impact. That's just not true. Again, please reference performance numbers in the future. -- AnthonyLiguori --70.112.17.156 05:14, 21 September 2006 (UTC)

[edit] Do not place external links to web consoles for Xen

Wikipedia has a strict external linking policy. See Also should only contain interwiki links. External links should only be for official sites or deeper information that wouldn't be appropriate for the article itself.

[edit] Enlightenments

The main section says "ported", aka "enlightened". They really aren't the same thing. Enlightments are a term invented by Microsoft to describe their paravirtualization technique. Viridian actually relies on having full virtualization capabilities in the hardware and essentially always runs a guest in full virtual mode. However, to acheive performance similar to a true paravirtual hypervisor, they "enlighten" the Operating System such that things that would normally be a bottleneck under full virtualization (for instance, a mass update of a PTE), can be batched. This really isn't what Xen currently does.

Retrieved from "http://en.wikipedia.org../../../x/e/n/Talk%7EXen_0d91.html"