Slashdot Mirror
Microsoft to Work with Xen on Virtualization
suso writes "Microsoft has released a statement to the press, saying that they are to work with Xensource on making Windows Server work with Xen through Microsoft's own hypervisor technology." Coverage available from Reuters as well. From that article: " As a result of the collaboration, the next version of Windows Server, code-named 'Longhorn,' will provide customers with a virtualisation system that promises to help run both Windows and Linux on the same machine more cost-effectively. Microsoft said it expects to conduct a public trial of Windows Server virtualisation by the end of this year and to release a commercial version of the software within 180 days of the date when Windows Server 'Longhorn' is released. Microsoft aims to release 'Longhorn' by the end of 2007, it said."
No longhorn is the codename for Vista. Blackcomb is the codename for Vienna (Vista server) if I remember correctly. The codename for Win95 was chicago, also if I remember correctly.
My company is currently working on this. We currently have, in development, a linux box going with Postgresql, and VMWare, win2k3 running in VMware. The reason?
Our application requires 2 things.
1: Postgresql (Performs way better on linux than windows for us)
2: IIS ((ASP.NET), we've tested mono... "it isn't there yet")
This allows us to ship linux and windows on the same box, and get the advantages of both, without shipping two boxes.
Can all fish swim?
It seems like the logical direction of Xen's progression.
d ex.html
From the creator of Xen's website "A port of Windows XP was developed for an earlier version of Xen, but is not available for release due to licence restrictions"
http://www.cl.cam.ac.uk/Research/SRG/netos/xen/in
The only thing standing in the way of Xen's running Microsoft's products is Microsoft.
There's a pdf on the Xensource website with information on how to install Windows Server 2003 right now. You do need a VT enabled Intel or AMD-V system though:
xensource
Advertising Software
Not true. In order for XEN to run an OS, that OS has to have XEN support in the kernel even on VT-enabled hardware.
No, that does not appear to be what that press release is saying.
It's saying that Microsoft will be able to run Xen enabled guests on Windows Server. IOW... They are acting as the Xen hypervisor and can run canned Xen images. I don't see anything in that announcement that gives me any encouragement to believe you will be able to run Windows under the native Xen hypervisor. They're not talking about "running Xen" only "running Xen enabled guests under the Microsoft virtualization". Different critter.
In fact, reading that release, they seem to have gone to great pains to word it to exclude running Windows as a guest under the Xen hypervisor. It's all spun very carefully, there.
You probably have some insider information but everything I read so far suggested that with hardware virtualization support Xen can run unmodified Windows. See Xen passes Windows milestone for example. Another example, Novell follows Red Hat with Xen announcement which says "But with hardware features in Intel processors today and AMD processors due in coming months, Xen will be able to run other operating systems as well, including Microsoft Windows.".
*blink*. VMware could run Windows when it first started shipping products eight years ago. It's Xen that has only been able to run Windows in the past few months, with VT chips. And Xen has conspicuously NOT published any benchmarks of that configuration - which leads me to believe performance sucks. Are you talking about paravirtualization? Fun fact: after VMware released their VMI patches to LKML, the very next release of Xen patches included several interfaces cloned directly from the VMware interface.
A witty [sig] proves nothing. --Voltaire
http://en.wikipedia.org/wiki/Trusted_computing
The big difference between ESX and virtual server is in the VM management - with VirtualCenter and a SAN, you can move running VMs between hosts, to redistribute load, or for maintenance. Granted, VirtualCenter is yet another piece of expensive software to buy. I'd guess there are still advantages to running ESX on its own, compared to Virtual Server, in that it runs on bare hardware, with (presumably) less overhead than running a full version of windows with VMs on top of it. ESX is also braindead simple to install and admin.
No, you're wrong. We have W2k3 running inside a Xen-enabled FC5 kernel. Try the instructions at http://www.xensource.com/files/xen_install_windows .pdf.
That's one of the major reasons why VT hardware support is so exciting.
Having been there and seen what was going on, some information, mostly public and available.
Xen can run Windows, on VT hardware, with a fair bit of fuss and squabble, and has been able to do so for quite a while. Performance isn't something I've looked at specifically (we were much more concerned with stability), but in theory it's going to be somewhat worse than a pure Xen paravirtualized system (the "traditional" Xen model), but better than virtualization as practiced by VMWare. What you do in hardware should be faster than what you do in software. Honestly that's just theory and I don't have numbers. Mind, VMWare are probably going to be using the same VT tricks in their next-gen stuff too.
Note too the difference in terms: Xen is doing some fundamentally different stuff than VMWare, and either needs a modified OS kernel (a trick that's also been done, cloak and dagger-like, or so I'm told), or hardware support (eg: Intel and AMD's VT/Pacifica technology). VMWare essentially provides a virtualized hardware platform of a stock configuration (Xen pretty much shows the guest what you've got on bare iron). For VMWare's workstation and GSX (now VMWare Server) products, bare-iron support is provided by the host OS. For ESX, as I understand, it's VMWare itself that has to provide drivers, and the HCL is pretty slim (current gig has run into this issue several times already). One of the beauties of Xen is that virtually (;-) all hardware support is provided by the Dom0 instance, or for all intents and purposes: Linux's driver support. Which is pretty darned good.
Xen's performance running non-VT is damned impressive. You'll note that VMWare expressly prohibit you from disclosing performance numbers with ESX. Workstation/GSX are typically a 30-40% hit, varying with workload. Xen, running standard Java benchmarking, trotted in at 2-4% consistently, adjusting for total available memory (Dom0 overhead isn't available to the benchmark, ~96-128 MB typically), plus a hair for the Xen hypervisor itself. This being both single instance and multiple guests in aggregate. Pretty good stuff.
I'm sorry to see XenSource getting close to the Vole, but do wish the good folks there well.