lkml.org 
[lkml]   [1998]   [Oct]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: linux-kernel-digest V1 #2628
Date
From
>From: Nathan Hand <nathanh@chirp.com.au>

>However I don't think modularised schedulers are nearly as trivial
>as you and the poster before you are suggesting.

You probably haven't heard of Vassal. I hadn't until yesterday. See

http://research.microsoft.com/~mbj/papers/UsenixNT98/vassal.html

This added, and I quote:

The Vassal changes made to the Windows NT kernel to support
multi-policy scheduling added 188 lines of C code, added 61 assembly
instructions, and replaced 6 assembly instructions.

The proof-of-concept external scheduler described earlier required
only 116 lines of C code and no assembly language. We believe these
are extremely low code size numbers for the increased functionality
that we achieved.

I don't believe that modular schedulers are trivial to add, but they
certainly don't have to involve a lot of code. Also, having screwed
around with the Mach scheduler to make it do something utterly
different than was intended originally, I don't think of problems like
this as too hard. This is perhaps impetuous. Also, Mach tended toward
the design-heavy end of the OS spectrum. Linux is extremely clean, but
not always design-clean. Linus is a good decision maker on the
elegance-versus-performance tradeoff, and the Linux scheduling code is
not quite as packaged as the stuff in Mach or NT. This is a good thing
for almost every reason, except when you want to do something very new.

>From: Richard Gooch <rgooch@atnf.csiro.au>

>Agreed. The proposal I saw looks really fancy and powerful, but I
>suspect it's awfully bloated. Just to have the basic framework in
>place would probably make the core scheduler 10 times slower than a
>simple scheduler (like Linux).

Well, in the Vassal case, having the hooks in the scheduler had
essentially *NO* effect (lost in the statistical noise).

With an external scheduler loaded, it slowed the scheduler down by
about 8%. The NT scheduler isn't very zippy compared to the Linux one
anyway, so its hard to predict which way this figure would go. Also,
the Vassal implementation was just a prototype, so it could get
better, or it could (unlikely) get worse.

--p

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:44    [W:0.033 / U:0.436 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site