Re: Kernel testing

Oliver Xymoron (
Fri, 11 Apr 1997 08:29:04 -0500 (CDT)

On Fri, 11 Apr 1997, Seth M. Landsman wrote:

> On Thu, 10 Apr 1997, Oliver Xymoron wrote:
> > A formalized testing process would be great - I'm envisioning a collection
> > of people with spare machines, each with slightly different configs,
> > downloading a kernel patch and a test suite patch, building the kernel,
> > rebooting, and then building and running an automated test suite for a day
> > or two ("make test"). The test suite might consist of a giant Perl script
> > coordinating a bunch of programs like crashme, lmbench, TCP/IP exercisers,
> > scripts to abuse various file systems and devices, and so on. It could
> > even include automatic reporting to a central database of some sort.
> > Hopefully with enough participants, a decent subset of possible kernel and
> > hardware configurations and features could be checked and we'd have some
> > data to point to for people with reliability concerns ("This kernel has
> > been tested on 150 different configurations for a total of 7200 hours
> > without crashing, enjoy.").
> FWIW, this is a great idea, and I'll help (not volunteer to
> coordinate, but help) write this thing if there is an interest ... The
> one caveat I see is that we should create a test suite that will run in
> approx. 6 hours. I would be (and I think so would others) willing to run
> this thing on my office machine, if I can set it going when I leave in the
> evening and have my machine back when I come in the next morning ...

I think the time is very flexible - I was thinking the bulk of the time
would be spent on stress testing like crashme rather than functionality
testing. Even a very complete set of functionality tests would be well
less than 6 hours, I'd think, because any single "does this syscall
return the right value?" test is going to be nearly instantaneous.

 "Love the dolphins," she advised him. "Write by W.A.S.T.E.."