In <1992Aug27.135703.9312@crd.ge.com> davidsen@ariel.crd.GE.COM writes:

quoted text
>    Here comes that idea again... The first o/s I helped write ran about
>  2/3 of the kernel in user mode, with user programs mapped into the
>  addressing space. Multics was using rings to get some of the same
>  things you get with setuid(), namely a limited set of privileges. GCOS
>  used a multi-threaded kernel (sort of) with multiple processors all
>  scampering around inside waving flags at one another. It even had
>  almost lightweight processes to handle i/o interrupts in user space.
>    That was mid 60's and it's interesting that the idea of monolithic
>  kernel is once again drifting out of vogue. Unfortunately I don't think

While the Linux kernel does its job well, its being monolithic is a
problem since all of the parts are interdependent, and to comprehend
the work of one of them, you have to know the whole system. This may
work for Linux but is unacceptable for bigger systems.

quoted text
>  the multiserver is right direction, since everything else in computers
>  is headed for less compleity rather than more. Multi-server is the CISC
>  of software, a sort of hypercube of processes rather than processors.
>
>  I'm all in favor of modularity (look at some of my net code), but I am
>  not convinced that this is the best way to get there.

Depends on what you want to achieve. A true distributed system could
transparently allocate resources of different machines to a job.
Whether you want this is another question.

quoted text
>    I like the Linux RISC-like approach, do only a few things, but very
>  well and very fast. Build the complex functions out of sequences of
>  simple operations. To me this means simple kernel calls and the library
>  providing the complex stuff.

I completely agree, and I wonder why this issue now is raised again by
Linux where this was the design principle of the original UNIX 20
years ago.  And now UNIX is a huge giant with a lot of memory donuts
:-) needed to feed him.

But for this approach you need at least shared libraries to avoid a
large-scale memory waste. And it comes down to the Amiga which has NO
real OS kernel, where ALL is shared libraries. Seems to me one of the
most well-designed OSs that ever existed, but been undervalued mostly
because of hardware and marketing problems :-(
(one of them being hardware dependence...)

quoted text
>    Don't take this as a rejection of multi-server by me, I'm unconvinced
>  rather than convinced against. Sort of a software agnostic.

Again: Computers are NOT the right place to practice religion. :-)
Better an agnostic than determinedly believing in something that can
possibly be proven wrong. :-)

MfG,
        Olaf
-- 
Olaf Titz - comp.sc.student - Univ of Karlsruhe - s_titz@iravcl.ira.uka.de -
uknf@dkauni2.bitnet - praetorius@irc - +49-721-60439 - did i forget something?
Der gr"une Punkt ist halt genauso sinnvoll wie ne T"UV-Plakette
 auf nem Schrotthaufen. - Thomas Volkmar Worm