Re: oskit-mach

[Roland McGrath]

> I have a system here that I want to run headless - I was thinking
> since I don't need the console on it, it might be cool to run
> oskit-mach on it.

Cool, yes.  But off hand I would still recommend gnumach rather than
oskit-mach for anyone who is not actively interested in debugging an
unstable microkernel.  OTOH, with the serial console and a power controller
it is easy enough to set up some expect scripts or suchlike on the machine
on the other end of the serial port to reboot it automagically when it
crashes.  (BTW, I recommend conserver (free software from conserver.com) or
something like it for handy logging of the console output and easy
multiplexing of remote access to it.)

I don't have any particular suggestions on exercising the kernel.  Most
bugs have been found by doing big compiles.  But that's because the system
has only been used by this here bunch of boring hackers, so compiling is
the only thing that we've really tried a whole lot (sure as hell the only
thing I ever do!).  A native build, install of libc and the hurd is good.

The main things I can think of to stress the kernel per se are basically
inducing paging (all hurd file i/o is paging), and inducing interrupt and
switching load.  So, do two things at once and use the network heavily at
the same time, for example.