[plug] the GNOME panel that just won't die

[Cameron Patrick]

Craig Ringer wrote:

| Any opinions or comments on the "virtual server on server using UML"
| approach?

The main caveat that I can see is performance.  For computational
stuff, UML gets within a couple of percent of the host system's speed,
but for anything remotely IO bound, UML is slow.  You can make UML
faster by applying the "SKAS" patch to the host kernel.  Still, since
a terminal server is likely to be doing lots of disc and network IO
you'd see quite a performance hit.  The best way to find out just how
much slower it is would be to try it -- UML is quite simple to set up
if you're used to compiling kernels.

I'd imagine that COW filesystems would come in handy for testing
things too.

Another comment: your server is SMP, yes?  As far as I know, UML
currently doesn't do SMP, although you can run two or more UML
instances on an SMP machine and take advantages of the extra CPUs that
way.

Cameron.