Re: Can someone explain please.

From: Andrew Valencia <vandys_at_nospam.org>
Date: Tue Aug 08 1995 - 08:07:33 PDT

[Vanier Kethireddy <vanier@aurora.carleton.ca> writes:]

> What is the reason for defining system calls (other than message
>passing primitives) at the kernel level, say for example fork or tfork.
>Why can't a process manager handle this job.

For that matter, why have messaging kernel calls when you can have a message
manager?

What I found, first in reading the available literature on microkernels, and
then in writing one, was that the goal of a microkernel is NOT to see how
much you can do outside of the kernel. Rather, the goal is to have the
microkernel offer a nucleus of services, such that the vast majority of
system enhancements you will make afterwards can be done outside of the
kernel.

Thus, the activity of defining kernel services was to ask "where does this
lead?". Not "is there any possible way I could do this outside the
kernel?". For instance, an external process manager *could* be
accomplished. However, you need to examine what it bought you (its failure
still kills your system, probably), what it cost you (performance issues),
and whether in balance it was worth breaking it out from the kernel.

> I would appriciate if some one can explain to me w.r.t boundry
>line between microkernel and the process manager i.e what exactly they
>are defined to do under vsta. At this point I do not see any difference
>between micorkernel and the process manager.

Right. I put processes, memory, and messaging in the kernel. Other very
worthy operating systems (QNX, in particular) have put the process
management outside. There are no "right" answers, just a continuum of
possible design choices.

> Am I thinking in the proper direction? If not can someone guide me
>please, I might be overlooking other issues.

While reading the process handling code, you might want to consider the
aspects of symmetric multiprocessor handling, as well as preemption. How
would these work outside the kernel? Pay particular attention to latency
and scalability.

                                                Regards,
                                                Andy
Received on Tue Aug 8 05:54:43 1995

This archive was generated by hypermail 2.1.8 : Thu Sep 22 2005 - 15:12:27 PDT