Re: [PATCH,RFC] fix o(1) handling of threads

[Ed Tomlinson <tomlins@cam.org>]

On January 2, 2003 07:22 pm, Ingo Molnar wrote:
> On Wed, 1 Jan 2003, Ed Tomlinson wrote:
> > Here is the scheduler-tunables patch updated to include USER_PENALTY and
> > THREAD_PENANTY.  This on top of ptg_B0.
>
> there's no way we'll make the scheduler internal constants tunable in such
> a wide range. Such a patch has been submitted a couple of months ago
> already. I do use something like that to test tunings, but it's definitely
> not something we want to make tunable directly in the stock kernel.

Nor would I advocate doing so.  I added two 'constants' I wanted to
be able to test them so I updated Robert's patch...  Two questions
for you.

1. Do you have any comments/suggestsion on the ptg_B0 patch?

2. I have been playing with using user and thread penalties together.
- they often interact badly.  Using just one works very well.  This
can be fixed - but gets messy.  Alternately, I am thinking about 
implementing per user policies.  ie.

	a. govern thread groups
	b. govern all threads, ignoring groups, for a user
	c. govern processes for a user

This can be done cleanly.  Would something along the lines of sys_nice
be the way to implement the kernel side of the user interface to this?

TIA,
Ed Tomlinson