Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!unido!marco!krabat!leo
From: leo@krabat.marco.de (Matthias Pfaller)
Newsgroups: comp.os.minix
Subject: Re: Most requested features in MINIX...
Message-ID: <743@krabat.marco.de>
Date: 5 Feb 91 08:52:01 GMT
References: <43516@nigel.ee.udel.edu> <8930@star.cs.vu.nl>
Organization: marco GmbH, 8047 Karlsfeld, West-Germany
Lines: 35

In article <8930@star.cs.vu.nl>, ast@cs.vu.nl (Andy Tanenbaum) writes:
> >16. Better scheduling algorithms.
> It is a single user machine.  Round robin should be good enough for that.

You should know that this simply is not true;
On page 80 of "Operating systems: Design and Implementation" you mention some
criteria what a good scheduler should try to achive. For a single user
machine I think that at least point 1 and 3 are interesting.
1. Fairness: make sure each process gets its fair shar of the CPU
	This is NOT true for the current scheduling algorithm of minix.
	Compute-bound processes have a higher preference then IO-bound
	processes. This is because if a IO-bound process gives up the CPU
	to make a call to the fs. Now the CPU-bound process consumes his
	timeslice complete, even if the systemcall of the IO-bound process
	has allready finished; So the CPU-bound process gets 99% of the
	CPU-Time.

This makes point 3 impossible:
3. Response time: minimize response time for interactive users.

One of the features of minix is Multitasking/programing; So if I compile
in the background, I wish to have a acceptable responsetime in the foreground.
I have read all the articles about multithreading fs... and I thought:
Ok, the bad response time is because all these fs-calls can't be done in
parallel. But then I build Kai-Uwe's priority-driven scheduler into my
minix 1.5-kernel. From this point of time I never even wasted ONE thought for
a multi-threading fs; The responsetime is now acceptable; I can compile
in the Background and let an editor running in the foreground; I almost allways
get immediate response to key-presses.

I think a better scheduling algorithm would be a real winn for minix and
1. It's a real small change to the kernel (only a few lines)
2. You would hear the 'why is fs not multi-threaded ...' much less

	Matthias Pfaller (leo @ marco.de)