Hi,
Although I was very happy with xen, it seems xen is losing out and kvm is the way to go. I had two migration questions that I couldn't find answers for.
1) Does kvm also run with para_virt (eg without VT capable cpu, but also just to be slightly more effecient). virt-manager only gave me a qemu option on my old amd64 cpu.
2) I boot all my xens using kernel+initrd outside a rootfs image, instead of a bootable virtual disk. Is that possible with kvm, or will I be stuck creating bootable disk images?
Paul
Paul Wouters schrieb:
- Does kvm also run with para_virt (eg without VT capable cpu, but also just to be slightly more effecient). virt-manager only gave me a qemu option on my old amd64 cpu.
While KVM can run paravirtualized Xen guests using xenner (still in development!), KVM needs always virtualization support in the CPU.
fs
Paul Wouters wrote:
- I boot all my xens using kernel+initrd outside a rootfs image, instead of a bootable virtual disk. Is that possible with kvm, or will I be stuck creating bootable disk images?
qemu provides -kernel and -initrd flags for this; however I don't know if virt-manager supports this.
I find that a kernel+initrd inside the guest disk image is much more maintainable; for example 'yum update kernel' will update your guest kernel rather than having to copy files around.
On Sat, 28 Jun 2008, Avi Kivity wrote:
qemu provides -kernel and -initrd flags for this; however I don't know if virt-manager supports this.
So with kvm i am still using qemu?
I find that a kernel+initrd inside the guest disk image is much more maintainable; for example 'yum update kernel' will update your guest kernel rather than having to copy files around.
I find it much easier to mount fs images, then to go hunt for offset's for mount in disk images. Also, during the xen2->xen3 and the non-PAE to PAE migrations, I liked my kernel and initrd separately :)
Paul
Paul Wouters wrote:
On Sat, 28 Jun 2008, Avi Kivity wrote:
qemu provides -kernel and -initrd flags for this; however I don't know if virt-manager supports this.
So with kvm i am still using qemu?
Yes; qemu provides emulation for the disk, network, and display devices, etc.
I find that a kernel+initrd inside the guest disk image is much more maintainable; for example 'yum update kernel' will update your guest kernel rather than having to copy files around.
I find it much easier to mount fs images, then to go hunt for offset's for mount in disk images.
Linux 2.6.26 supports partitionable loops devices, so you can loopback mount your disk image and then mount filesystems off /dev/loop0p1, /dev/loop0p2, etc.
Also, during the xen2->xen3 and the non-PAE to PAE migrations, I liked my kernel and initrd separately :)
kvm supports pae and nonpae regardless of how the host kernel was compiled, so you can happily run an pae guest on top of a nonpae host.
On Sat, Jun 28, 2008 at 01:02:46AM -0400, Paul Wouters wrote:
On Sat, 28 Jun 2008, Avi Kivity wrote:
qemu provides -kernel and -initrd flags for this; however I don't know if virt-manager supports this.
So with kvm i am still using qemu?
Sort of. KVM uses a fork of qemu, although the two code bases seem to be regularly synched up. Anyway, the command line arguments are broadly similar.
Remember that virt-manager is using libvirt, and libvirt hides the qemu command line behind a configuration file (and also hides any differences): http://libvirt.org/formatdomain.html
Rich.
On Fri, Jun 27, 2008 at 05:56:37PM -0400, Paul Wouters wrote:
Although I was very happy with xen, it seems xen is losing out and kvm is the way to go. I had two migration questions that I couldn't find answers for.
- Does kvm also run with para_virt (eg without VT capable cpu, but also just to be slightly more effecient). virt-manager only gave me a qemu option on my old amd64 cpu.
KVM requires a hardware virt-capable CPU. These are quite common, (if your CPU was made within the last two years or so) but note there are at least two major 'generations' of virt technology in CPUs. In particular very new CPUs from both Intel and AMD support some form of virtualized page tables, which can be very beneficial. Intel describes them here:
http://www.intel.com/technology/itj/2006/v10i3/1-hardware/8-virtualization-f...
[Question: does EPT / NPT only affect 32 bit?]
The second area where Xen had a traditional advantage was in paravirtualizing disk, network and a few other devices. Linux now supports something called virtio, a standard for virtualized disk and network devices, and KVM (and others) support this. Here's some more info about virtio:
http://lwn.net/Articles/239238/
As for your problem with virt-manager, can you please tell us what it says in /root/.virt-manager.log and the output of 'virsh -c qemu:///system capabilities' command, and the contents of /proc/cpuinfo.
- I boot all my xens using kernel+initrd outside a rootfs image, instead of a bootable virtual disk. Is that possible with kvm, or will I be stuck creating bootable disk images?
KVM has some command line options to do this, namely -kernel and -initrd.
Rich.
Richard W.M. Jones wrote:
On Fri, Jun 27, 2008 at 05:56:37PM -0400, Paul Wouters wrote:
Although I was very happy with xen, it seems xen is losing out and kvm is the way to go. I had two migration questions that I couldn't find answers for.
- Does kvm also run with para_virt (eg without VT capable cpu, but also just to be slightly more effecient). virt-manager only gave me a qemu option on my old amd64 cpu.
KVM requires a hardware virt-capable CPU. These are quite common, (if your CPU was made within the last two years or so) but note there are at least two major 'generations' of virt technology in CPUs. In particular very new CPUs from both Intel and AMD support some form of virtualized page tables, which can be very beneficial. Intel describes them here:
http://www.intel.com/technology/itj/2006/v10i3/1-hardware/8-virtualization-f...
[Question: does EPT / NPT only affect 32 bit?]
No, it's applicable to both 32- and 64- bit guests.
The KVM NPT implementation does not work for 32-bit nonpae hosts (it will fall back to not using NPT), so if you have a recent AMD processor and you are using a 32-bit installation, use the pae kernel for best performance.
The kvm NPT and EPT implementations were introduced in Linux 2.6.26.
-
Avi Kivity -
Felix Schwarz -
Paul Wouters -
Richard W.M. Jones