Some KVM developments [LWN.net]

Some KVM developments

Posted Jan 13, 2007 19:47 UTC (Sat) by nlucas (subscriber, #33793)
Parent article: Some KVM developments

I always thought the KVM way was better than the Xen way, since I had contact with the colinux way of doing things. With the added support for paravirtualization, it should not be long until KVM will also support CPU's without the VMX hardware support, which will make many people wonder why they need Xen.

Paravirtualization will not make you run Windows on it if your CPU doesn't have hardware support, but will be enough to test and run different Linux and other open source systems with much better performance than with Qemu (for many things, users will not notice the difference).

It's even conceivable that a KVM application could do it in a transparent way to the user, and if the CPU doesn't support VMX it will fallback to using Qemu emulation to run the native system image (as KVM also needs Qemu support for emulating some hardware). So KVM will behave as the Qemu hardware accelerator module (the Kqemu closed source driver).

to post comments

Some KVM developments

Posted Jan 17, 2007 2:23 UTC (Wed) by ttfkam (guest, #29791) [Link]

It also opens up quite a few avenues that simply weren't possible before. For example:

Combined with the LinuxBIOS project, this allows a "bare" PC to be bootable while still allowing for Windows to be loaded; no longer either/or.

Dual booting is even easier as both OSes would be under the control of the hypervisor. With enough RAM and disk space, it would literally be two (or more) computers in one.

Sharing data between the OSes would no longer be dependent upon Linux support for NTFS -- or any other Windows or Mac filesystem -- but rather SMB/CIFS (read: Samba) access in both directions: Linux->Windows, Windows->Linux, Linux->OSX, and OSX->Linux.

Expanded use of kernel-built-in paravirtualization means that hardware drivers are handled at the hypervisor level while apps and config are at the guest level. Backup and restoration of servers cease to be hardware configuration-dependent. Having this is the vanilla kernel raises the possibility of a uniform backup utility for any and all flavors/distributions.

Syslog to the hypervisor by default so that even if a guest is compromised, the logs cannot be. (Or at least the difficulty in altering logs raises dramatically.)

That's off the top of my head. Any other bright ideas to share?

Some KVM developments

Posted Feb 5, 2007 19:12 UTC (Mon) by mmarq (guest, #2332) [Link]

"" With the added support for paravirtualization, it should not be long until KVM will also support CPU's without the VMX hardware support, which will make many people wonder why they need Xen. ""

Well in the datacenter, the possibilities to *migrate* VMs on failures seems to me to be invaluable. The industry is converging on the x86_64, so the ability to emulate... MIPS or POWER; does not seem to me, to be a very hard to pass thing.