Subject: Re: Floating point in the kernel
To: Curt Sampson <cjs@portal.ca>
From: Stefan Grefen <grefen@hprc.tandem.com>
List: tech-kern
Date: 09/20/1998 12:06:10
In message <Pine.NEB.4.02.9809181415470.10288-100000@ascetic.portal.ca>  Curt Sampson wrote:
> 
> It sounds to me as if we're going the wrong direction for this,
> and what we really want is some way to give the userland process
> enough priority that it's going to get the time it needs even on
> a heavily loaded system.

YES.

For something like this we need a minimum 'real-time-style' capability.
This can be either done by a niceness-threshhold after which the 
process/thread isn't part of scheduler priority changes any more,
or a kind of realtime scheduler ....

Doing a nice -n -40 softmodem, would run the softmodem always when 
its runnable (under any load). If it needs interrupts blocked ... then 
it shouldn't be anywere, because it would degrade performance of the
whole system too much.

The interrupt problem should be less, when we get kernel-threads, as the
HW-interrupts will do less work then.

The niceness-threshhold is easy to implmement and has no serious side-effects,
(unless the super-user uses it to boost his kernel compiles :-)))).

(Existing example Convex (Conevex OS since 8.0  ...)

Stefan

> 
> cjs
> --
> Curt Sampson  <cjs@portal.ca>  604-257-9400    De gustibus, aut bene aut nihil.
> Any opinions expressed are mine and mine alone.
> The most widely ported operating system in the world: http://www.netbsd.org
> 

--
Stefan Grefen                                Tandem Computers Europe Inc.
grefen@hprc.tandem.com                       High Performance Research Center
 --- Hacking's just another word for nothing left to kludge. ---