On Thu, Apr 06, 2000 at 05:17:36PM -0700, Brent Fulgham wrote:
>
> > I thought the Mach 4 base was supposed to be very good (with
> > kernel RPC transactions taking less time than traditional BSD
> > ontext mode / function call processing).
> >
> > Have any suggestions on what would be better to port to? Would this
> > depend on implementing libmom fully?
> >
> Well, libmom is pretty much dead at the moment. And I don't know
> that Mach is so bad that we can't produce a decent system on top
> of it. I mean, no one really complained about NeXT and I haven't
> heard too much about Mac OS X being a slow dog.
>
> As for performance, I'm thinking in terms of something like Fiasco
> (http://os.inf.tu-dresden.de/fiasco/faq.html), but it's structure
> might not be suitable for use with the Hurd (at least as it now
> sits). And who knows if the speed increase would be noticable. Most
> of our speed issues are probably caused by bottlenecks in other
> places right now.
I do not know what Darwin does, however, the way that OSF (aka, Tru64)
stays fast is two fold: `in process' message handling, minimizing context
switches; and avoiding the server concept that the hurd embraces.
What does this mean? Here is a scenario:
To access a file on OSF is still really fast because they decided to stay
with a traditional filesystem implementation: The application traps to the
kernel which checks the permissions, reads the data (possibly blocking)
and finally returns to the user application. In the hurd, the application
traps to the kernel, which in turn needs to call the auth, ext2fs and
storeio translators/servers. In turn, each of these traps to the kernel then
escapes back into user space to process their message. Finally, they can trap
back into the kernel to hand off the results and return that to the user
space application. This all translates to a lot of memory copying (from all
of the messages) and a slew of context switches, which as we all, are by
far the most expensive operation in any operating systems and even more so
on a CISC chip, the only architecture that the currently hurd runs.
In conclusion, being an idealist can be very slow, but also, very elegent.
-Neal
--
----------------------------------------------------------------------------
Neal Walfield [EMAIL PROTECTED]
UMass Lowell - Fox 1512 Phone: 978-934-5347
Fax: 603-415-3645
Love is the triumph of imagination over intelligence.
-- H. L. Mencken
