Currently we have assorted development build/test boxes. The target Application is mainly written in Python/gcc, uses postgres and has 2 identical USB devices attached.
The principal build OSes are RHEL, FreeBSD & XP on i686. The App needs to be regularly built and tested on a couple of releases of each OS.
(Maybe the next step would be to test on bonus OSes/release/CPUs eg Fedora, SuSE, Debian, Solaris and Vista, for both 32bit and x86-64 hardware, maybe even PPC.)
I was hoping that I could simply copy the existing Filesystems directly onto their own Logical Volume of a Visualization Server (Xen or VMWare), boot the VM's and use the existing test suits.
Then every day we could then revert the target OSes Logical Volume to it's original state, then run the build and test scripts.
One VM/LV per test server, running on one Visualization box seems to be a good idea, but I have encountered some problems.
Problems encountered so far are:
VMWARE
Handles BIOS/Hardware better, doesn't like a VM on VLM
- Wont boot a Virtual Machine off a Logical volume.
- FileSystems must be converted to VMFS for VMWare snapshots.
XEN
Logical Volume snap shots work perfect, and LVs can be extended.
- problems visualizing XP & FreeBSD
- problems dealing with raw USB devices.
- There are also issues with X11 hanging.
I have not tried any other visualization solutions. {Wikipedia virtualization software}
Are there any other choices or paths I should consider?
Suggestions, working examples, white-papers and/or FAQ on such test systems welcome.
Ben