Joshua Penix wrote: > On Dec 17, 2007, at 1:26 AM, James G. Sack (jim) wrote: > >> I guess I'm interpreting aol correctly that (xen) full virtualization >> just plain doesn't work for Windows guests. That's disappointing -- the >> promise was too good to be true, I suppose. > > > Sorry, I was a little unclear. Assuming version 3.x of Xen and > available VT/AMD-V CPU extensions, Xen's full virtualization, even the > open source version, will work for Windows guests. It just won't work > terribly *well*... that's where the commerical XenServer's drivers come > in to help. Basically, the network and disk drivers that the standard > HVM device model emulates and exposes to Windows are ridiculously slow. > I haven't benchmarked it myself, but I've seen reports of 25-50% of > native performance. Now, for some people who just need to pop open > Excel once a week on their Linux desktop, this may not matter. But in a > server environment where Windows is actually supposed to do some real > work, the additional drivers are almost required, hence my use of > XenServer Enterprise in the writeup. > > The situation is also confused a little more by commercial Linux > distributions that include Xen. RedHat is very clear that their Xen > support is only meant for virtualization of other Linux instances. > Windows may run, but don't call them to talk about it. > > Oracle's new virtualization solution, also Xen-based, holds a little > more hope. They actually list Windows as a supported OS, but explicitly > state that performance is going to suck. But looking at their FAQ, > there's this interesting sentence: "Oracle is developing paravirtualized > Windows drivers that will substantially improve the performance of > Windows on Oracle VM." And since Oracle's solution is currently a free > download for everyone (you can just choose whether or not to hand them > money for support), the key question is whether or not those drivers > will also be made available for free or if they'll start hiding them > behind the support wall. Time will tell! > > As a bonus... if you have a Xen machine with the right CPU and want to > quickly test out Windows support, here's a simple Xen config file that > should get you booted into setup. Just change the paths on the 'disk' > line to point at 1) an empty LVM partition; 2) and your Windows OS > image. You may also have to change the path to the 'hvmloader' and > 'qemu-dm' files, as this particular snippet comes from an Ubuntu system. > > ----- > kernel = "/usr/lib/xen-ioemu-3.1/boot/hvmloader" > builder="hvm" > memory = 512 > name = "w2k3-001" > disk = [ 'phy:/dev/Guests/w2k3-001,hda,w', > 'file:/images/Windows-2003-Server-R2-VLA-Disk-1.ISO,hdc:c > drom,r' ] > device_model = "/usr/lib/xen-ioemu-3.1/bin/qemu-dm" > boot="c" > vnc=1 > usbdevice='tablet' > vif = [ 'type=ioemu, bridge=xenbr1' ] > ----- > > Once you 'xen create' that config, connect VNC to localhost port 5900 > and you should see the Windows installer! When the install is done, > change 'boot="c"' to "d" and it will boot from the virtual hard drive > into the OS. >
OK, thanks for the update. When are you going to write your _Virtualization Recipes_ book? Regards, ..jim -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
