It seems like Jakob Bohm's cases A and B aren't too difficult technically, since both software emulation and KVM emulation are of the same architecture -- there is a well-defined state of the emulated system that doesn't depend on how it is being emulated. The biggest problem is ensuring that the live-migration system transmits and receives data in a way that is independent of how the guest is being emulated. But such a way can be found.
> From: Prashant Upadhyaya <prashant.upadhy...@aricent.com> > > Case A and B are what I am looking for. > I believe B is more likely for practical usecases for speed of guest > execution. > > I saw this news from 2008 and thought that years have passed, so > this must be possible now. But looks like this still hasn't matured. > > http://beta.slashdot.org/story/109745 To me, a more interesting question is why anyone cares. A software-emulated system is 10x slower than hardware, so the general ability to move a guest onto a host with a different CPU architecture is not very desirable from a practical point of view. People aren't going to want to move processes between CPU architectures very often, so in practice they should be able to shut down the process on one CPU and start it again on the other. (The application has to support that sort of thing anyway.) > Would love to hear comments on the above story from back then. Qemu-discuss seems to be no older than 2011 Oct (see http://lists.nongnu.org/archive/html/qemu-discuss/), but Qemu-devel goes back to 2003 Apr (http://lists.nongnu.org/archive/html/qemu-devel/). Dale Dale Worley -- Dave Friedman, legendary veteran of the exploitation market, proudly boasts "Hell, I made films that made 'Plan 9 from Outer Space' look like Citizen Kane." -- Calligan and Haber [http://www.imdb.com/name/nm0293354]
