Submitted by Xen 3.0 has been released. “We’ve been seeing good stability on the XenRT regression tests for the last couple of weeks, and the number of bug reports submitted to bugzilla have dropped right down.” Get it here. “Along with the usual binary install tarball, we’ve created a new live-iso demo CD, and some RPM packages for common linux distros.”
I expected such from Slashdot, but osnews.com should know better.
So, could someone kindly inform the rest of us, what Xen 3.0 is?
mario: XEN is a built in Virtual Emulator. Much like Vmware, but has been reproted as MUCH faster as it is closer to the metal (being built into the kernel)
XEN is Open Source, and is include in a couple Linux distors already, SUSE/Fedora, and this new version shows some real promise.
OSNEWS has covered a few stories on XEN, as it is a very cool new bunch of code.
It’s not an emulator. It’s a hypervisor. Parent should have just opened up a new tab and looked on Wikipedia. Xen is covered a lot here and not uncommon.
http://en.wikipedia.org/wiki/Xen
We have covered Xen quite a few times, and other than that, Google is only a few clicks away.
Next thing, I’m gonna have to add descriptions about what Windows or Linux is
. The space is limited, people.
“We have covered Xen quite a few times, and other than that, Google is only a few clicks away.
Next thing, I’m gonna have to add descriptions about what Windows or Linux is
. The space is limited, people.”
I’ll do it for you!
Windows is…
A 32-bit extension of
a 16-bit graphical shell of
an 8-bit operating system coded for
a 4-bit processor by
a two-bit company that can’t stand
one bit of competition.
Linux is…
The perpetual beta that results when a bunch of religious fanatics decide to write an operating system.
Is to have my Linux system, that already virtualizes serveral Linux systems able to also run Windows under Xen. That would be really great.
My understanding is your kernel has to be Xen patched to run under Xen right? If thats the case we are not going to see Windows support.
As long as your processor supports VT or Pacifica, you can run unmodified OSes under Xen. It’s not fully debugged yet, but then most people don’t have the hardware anyway.
Not at all what i submitted lol…not sure where that info came from lol..
Xen is a virtual machine monitor for x86 that supports execution of multiple guest operating systems with unprecedented levels of performance and resource isolation. Xen is Open Source software, released under the terms of the GNU General Public License. We have a fully functional ports of Linux 2.4 and 2.6 running over Xen, and regularly use it for running demanding applications like MySQL, Apache and PostgreSQL. Any Linux distribution (RedHat, SuSE, Debian, Mandrake) should run unmodified over the ported OS.
…until it supports Windows with an unpatched kernel. Come on guys, get the finger out! VMWare currently buys and sells Xen
It does run Windows – you just need the need VMX / SVM hardware from Intel / AMD. It’ll be available to the public shortly (in Intel’s case) and some time next year (in AMD’s case).
But the point of Xen is really to run OSes “natively” on a hypervisor. Since they understand that they’re not on a real machine, many of the performance problems of full virtualisation on x86 are eliminated.
Conversely, VMware is looking at paravirtualisation (patching Linux kernel to run with better performance on VMware) and will probably be using VMX / SVM to enhance their full-virt performance. We live in interesting times for virtualisation work!
Your comment is misleading!
No, Xen does not run Windows as it requires something which is not available NOW.
It may run Windows in the future, I agree with you, but it does not run Windows.
I am sure that with a PhD, you see the difference 😉
Rather trollish. There are a TON of uses for any version of Xen. The world DOES NOT revolve solely around Microsoft’s Windows. Compared to ESX/GSX’s costs, hoping Xen works out is a good thing. I am pretty excited to hear Xen progress anytime I can.
@Mark: Thanks man for the info. Xen is sweet, and keep up the great work!
-iGZo
Thanks OSNews. I look forward to hearing updates on Xen. Sounds like this is absolutely amazing stuff.
“Along with the usual binary install tarball, we’ve created a new
live-iso demo CD, and some RPM packages for common linux
distros.”
Even a live-cd demo! What don’t this guys do?
(I’d also love to hear more on integration with Vanderpool and Pacifica.)
(disclaimer: am a Xen dev, should have mentioned it in the other post sorry. If anyone disagrees my points, feel free to roast me for justification;-)
Xen 3.0 adds support for x86_64 and for PAE mode on x86_32 (older systems with > 4Gig of memory). It also adds a beta port for IA64, with ports to other architectures being still in progress.
Vanderpool support is already integrated and has been demonstrated (at Intel IDF) running Windows XP. This support needs further development to perform as well as we want – there are plans to rearchitect the full virtualisation IO model for a future release. Much of the logic (particularly for IO) will be common with Pacifica, so these changes will also benefit AMD users, after AMD release the Pacifica support patch they’ve been working on.
Regarding Xen-native paravirtualised OS support:
Native ports are able to achieve the maximal available performance and do not need virtualisation-aware hardware to run.
Right now, a Xen 3.0 aware patch for Linux 2.6 is available. We’re intending to get Xen-awareness checked into the mainline Linux kernel in the future. NetBSD 3 includes Xen 2.0 support but Xen 3.0 support is being worked on by those guys. FreeBSD 6 will support Xen 3.0 at some point in the future – most of the code is written but it needs updating for recent changes in both FreeBSD and Xen. The Plan 9 port is also being updated, and there have been rumblings of ports of both Minix and ReactOS.
Note that most of the porting is being worked on by community members, not the core Xen team.
I have a couple of question I hope you can answer.
As far as I understand, Xen requires Vanderpool or Pacifica technology. All Pacifica enabled CPU’s are 64-bits and does not require PAE. Is any of the Vanderpool enabled CPU’s 32-bit or is there some other explanation for implementing PAE?
I can understand that an Operating System can benefit from being aware of a hypervisor, but is necessary to have a modified kernel to run the OS?
As I understand, Xen still chunks a single machine into smaller independant pieces. This seems very wasteful to me. Consider openvirtuozzo (www.openvz.org), where each virtual host sees the machine as a whole. This is possible on the base that all the virtual hosts need all the hardware capabilites at the same time very rarely. So you can actually let each virtual host use the whole machine while putting just minimal guarantees on the cpu time slice and memory to it. Very neat imho.
There is nearly zero competition between Xen and OpenVirtuoso. Xen is hardware virtualization and OpenVirtuoso is OS virtualization. Xen more-or-less competes with VMWare and whatever Microsoft’s virtualization server is called, OpenVT compares roughly with BSD jails or Solaris Zones.
The grand vision being pitched to enterprise CIOs is a strategy combining hardware and OS virtualization: the former for availability/failover and utilization, the latter for isolation and load management.
Xen gets vastly more click-views than OpenVT at this point in time, which is the only shortcoming in the FOSS approach to virtualization. The FOSS hype machine needs to emphasize the fact that it offers a complete collection of virtualization tools.
Slicing up at the virtual machine level – even when its dynamically adjustable – is, as you say, a bit heavyweight for some applications but it also gives stronger isolation and the ability to suspend-resume and live-migrate guests OSes.
Personally, I’d vote for running an OS-level partitioning technology (like OpenVZ) on top of machine-level partitioning. I’d say it’s really a spectrum of technologies that can be combined, rather than an either / or. eg.
* The machine level virtualisation (e.g. Xen or another hypervisor) gives stronger isolation guarantees between certain critical servers, and allows you to live migrate running servers around your cluster.
* The OS-level partitioning allows you to securely isolate multiple services within the virtual machine.
* User accounts provide basic isolation within an OS level partition, providing the least isolation but being the easiest to set up.
XEN is the VMware Killer. So long VMware.
> XEN is the VMware Killer. So long VMware.
Not really. It is true that VMware will lose business because of Xen among the tech crowd. (I’m a happy VMware customer since version 3.1 that will likely switch when I upgrade to Vanderpool or Pacifica.)
But, notwithstanding, VMware has three key advantages over Xen:
1) It’s cross-platform — Linux, Windows, and soon the Mac. (Xen isn’t)
2) It’s virtual machines are cross platform. (Xen’s VMs aren’t)
3) It’s dead simple to set up. (Xen isn’t yet).
Point (3) will change with time, but points (1) and (2) aren’t going to change any time soon (if ever). VMware still has a solid place in business.
I just installed it for the first time, and it wasn’t very difficult to use.
Much faster than Qemu, but it doesn’t compete with VMWare right now. (I use Qemu for installing “scratch” installations of Debian for building packages in an isolated fashion, and VMWare for running Windows 2000 for browser testing with IE6).
I wrote a brief introduction here:
http://www.debian-administration.org/articles/304
Xen looks like it will be a strong product: features like live migration are what will move it from a developer/tester-only tool to something that regular SEs will want in production environments.
But, right now, VMWare is out there, commercially packaged and supported, and “just works”. With the Workstation version running $189US and the Viewer is free, that’s pretty much no cost at all for a company larger than 10 people.
Where I work we’re using it for porting to Linux and Solaris x86: and the ability to be running Windows (for Outlook) as well as Solaris x89 9 and 10 and RedHat and SUSE Enterprise all on one system, for every developer, is great! That one product is saving us almost 10x it’s license cost in additioanl hardware for testing and development.
That being said, of course we’re keeping our eyes on Zen. I’m just saying it’s not there yet. Xen vs. VMWare is like taking a knife to a gun fight (today)… but all the open source crowd is looking to make it some WMD’s
There is one thing I don´t understand, yet:
Should the host OS be patched – as there is always the mentioning of a Xen Patch to 2.6 – , should the guest be patched or should they both be patched?
Because – as here this discussoin already runs- if only the host OS should be patched why does a guest WinXP needs this special VT Intel processor?
This should only true if the guest also needs a patch!?
Can someone elaborate, please?
Thanks!
Both OSes must be patched, unless you have the hardware virtualization extensions.
> Your comment is misleading!
>
> No, Xen does not run Windows as it requires something which
> is not available NOW.
I’m sorry you think that but you’re misunderstanding what I said.
If you look back, you’ll notice I said “you just need the need VMX / SVM hardware from Intel / AMD. It’ll be available to the public shortly“. I don’t think anyone here has suggested you can buy the hardware now, although Intel has shipped to OEMs already.
Xen supports the hardware, that hardware physically exists (and has been available to developers for months), and Windows has been demonstrated running on that hardware. So yes, it does, run Windows, but not on your machine 😉
> I am sure that with a PhD, you see the difference 😉
Sorry for any confusion caused by the original post. I hope this clarifies what I meant. By the way, I don’t have a PhD yet – more’s the pity!
> As far as I understand, Xen requires Vanderpool or Pacifica
> technology. All Pacifica enabled CPU’s are 64-bits and does not
> require PAE. Is any of the Vanderpool enabled CPU’s 32-bit or
> is there some other explanation for implementing PAE?
Xen doesn’t require Vanderpool or Pacifica for running guests that are “Xen native”. The first OS port to Xen 3.0 was Linux 2.6 – the intention is to have support for native Xen operation merged into Linus’ tree at some point soonish. Support for NetBSD, FreeBSD and Plan 9 is currently being implemented (these all ran natively on Xen 2.0).
Vanderpool and Pacifica are only used for running *unmodified* guests, such as Windows, legacy versions of Linux, etc.
People have been running Xen for a couple of years on standard x86 hardware, with no special extensions to the hardware needed. The intention was to support normal x86_32 and also x86_64 but we turned out to have too many users with big memory (>4Gig) x86_32 machines – those users needed PAE support to make use of all their RAM.
So now we have:
* 64 bit Xen runs 64-bit “Xen native” guests. If the chip has Vanderpool / Pacifica it will also run 64-bit or 32-bit unmodified guests alongside.
* PAE bit Xen runs PAE “Xen native” guests
* 32 bit Xen runs 32-bit Xen native guests
If a 32-bit Vanderpool processor appears (I’m not clear on Intel’s roadmap) then the latter two would likely have the ability to run 32-bit unmodified guests.
>I can understand that an Operating System can benefit from
> being aware of a hypervisor, but is necessary to have a
> modified kernel to run the OS?
It’s not *strictly* necessary – VMware manage without. But doing this is less efficient since it allows fewer optimisations. It’s also technically very complex to virtualise unmodified OS kernels on non-Vanderpool / Pacifica hardware – that’s why Xen requires those if you want to also run unmodified guests.
Xen 3.0 is one of the most important software to be released in years.
But it is not a VMWare killer. For example, it won’t let me run Linux or FreeBSD under Windows 2000/XP. And it won’t let me run Windows 2000/XP under Linux.
VMWare and Xen are very different and unless you’ve used both, it is silly to compare them.
I believe Xen 3.0 is going to start a revolution in server hosting. It will enable consumers to have a hosted virtual Linux server with full root access at low-cost. For example, we’ll start seeing $15/month for a 96MB Linux with full root access the the prices will go down even more over time. By full root access, I mean the likes of which is provided by UML–not the pseudo roots provided by some of the other virtualization software used on servers these days.
Xen requires a modified kernel to run for now, but Red Hat doesn’t think it will take much time before Xen is added to the Linux kernel directly (couple months?).
http://www.osnews.com/story.php?news_id=12470
You will not be able to virtualize Windows on Xen without hardware support.
http://wiki.xensource.com/xenwiki/XenFaq#head-fcb85a149da66907086cc…
The ReactOS project is working on fully supporting Xen even without hardware support.
http://www.reactos.org/wiki/index.php/Xen_port
MS is planning their own virtualization solution that is separate from Xen. There isn’t much indication whether or not Xen will be able to host their solution without hardware support though.
http://www.microsoft.com/technet/community/chats/trans/windowsnet/0…
Any chance of Xen being ported to Darwin? If Apple adopts the Intel VT chips it could be quite powerful for them.
Because Xen’s a hypervisor, it runs at the lowest level in the stack – beneath all the OSes on the computer. Thus, it’s more a case of Darwin being ported to run on Xen.
There’s been occasional interest in a port of Xnu (the Darwin / MacOS X kernel) to run on Xen but I don’t have concrete information about an in-progress port at this time.
Thanks for providing the extra links! Very useful. Red Hat and SuSE are both shipping Xen-enabled kernels for optional install (in Fedora 4 and SuSE 9/10, respectively). It makes life much easier for them if Xen is in the mainline rather than being yet another patch.
> MS is planning their own virtualization solution that is
> separate from Xen. There isn’t much indication whether or
> not Xen will be able to host their solution without hardware
> support though.
The MS hypervisor is planned to require hardware virtualisation support (Vanderpool / Pacifica), although it is intended that certain performance critical operations will be optionally paravirtualised (to allow hypervisor-aware guests better performance).
There’s two aspects to this:
1) Can the MS hypervisor and related products run under Xen in a virtual machine?
My reading of the Vanderpool / Pacifica specs suggests that it’s technically possible to support nested hypervisors in this way. Obviously, nesting hypervisors is a performance penalty you’d rather not take, though 😉
2) Can Xen support the MS hypervisor interface, so that Windows will use the same optimisations under Xen that it uses under the MS hypervisor?
This is less clear. It’s certainly technically possible; it’s more a question of how open the MS hypervisor API specs are and what the restrictions on reimplementing them in other hypervisors.
Nb. VMware are also working on a third paravirtualisation API.
I’d expect to see native support for Xen in the mainline code of many of the major open source kernels within the next year or so. For NetBSD, Xen 2.0 support is already there, for FreeBSD the Xen porter has been given CVS commit access, for Linux the Xen 3.0 patches have already gone through multiple stages of review and restructuring, and should be submitted Real Soon Now for inclusion.
A major advance in terms of OS support is that the Xen 3.0 interface should be supported *for a long time* into the future. Once a kernel is ported to the Xen 3.0 interface, it will continue to run on future versions of Xen. Previously we’ve needed to change the interface periodically to accommodate new design decisions, so ports have needed updating to the new interfaces (this is why e.g. NetBSD is still using Xen 2.0).
I’d just like to point out that Xen isn’t Linux-only – it’s just that the core team who develop Xen also support the Linux patches so those come out first. Hopefully with the Xen 3.0 interface freeze, the previous OS ports for Xen 2.0 will be quickly updated to the new, frozen API.
For 1), NetBSD 3.0 offers Xen 2.0 support (“host” as well as guest). We are, however, unlikely to see Windows “host” support soon, if ever.
For 2) the virtual machine interface is OS independent and OSes other than Linux do run in Xen virtual machines – the only requirement is source code access so someone can port it (or Vanderpool / Pacifica hardware so that it can run unmodified).
For 3) – yeah it’s not as easy as we’d want 🙂 It’s a bit tricky because you have to install the hypervisor and a new kernel, rather than just an application. XenSource are now supplying installer scripts which help things a bit. However, I think the best way to make it easy is for distro installers to support Xen properly – this is getting there, SuSE seem to be doing well at this.
What are the differences betwen Linux-VServer and Xen ?
Xen virtualises at the level of the machine hardware, whereas Linux Vserver virtualises at the operating system interface.
When Xen, each virtual machine has a separate kernel, managing its own subset of the machine’s resources. With Vserver, virtual machines all run on the same kernel, which has extra hooks to hide activities in other virtual machines.
Xen’s approach allows different kernel versions or different OSes to run in each virtual machine. It also provides stronger isolation between virtual machines, and allows for virtual machines to be live-migrated to other hosts without stopping them. On virtualisation-aware hardware (Vanderpool or Pacifica) it’ll run Windows virtual machines.
Vserver’s approach is lighter weight, so it should have IO overhead for demanding workloads (e.g. high packet-rate networking) and allows greater statistical multiplexing of the machine’s resources. It’s convenient and simple to use.
Both of them solve different problems. Xen is analogous to VMware ESX server, Vservers is analogous so Solaris Zones or Virtuozzo.
Just got an email this morning from Parallels (http://www.parallels.com) stating that they just went GA, and that this version fully supports Intel VT architecture:
http://www.parallels.com/en/news/id,8345
Edited 2005-12-08 20:51