2002-07-02 17:09:08

by Robert Love

[permalink] [raw]
Subject: [PATCH] O(1) scheduler for 2.4.19-rc1

Available at

ftp://ftp.kernel.org/pub/linux/kernel/people/rml/sched/ingo-O1/sched-O1-rml-2.4.19-rc1-1.patch

and mirrors. Aside from the resync to 2.4.19-rc1, the following changes
are new since the last release (most all pulled from 2.5):

- reintroduce sync wake ups
- whitespace cleanup, trivial cleanups
- remove frozen lock and introduce new arch-specific
switch_mm() logic
- new rq_lock and rq_unlock methods
- wake_up optimization
- nr_uninterruptible optimization for count_active_tasks
- merge the task CPU affinity system calls
- sched_yield bugfix
- minor fixes

Compiles on x86 UP and SMP.

Since Ingo recently posted 2.4-ac resyncs, I will refrain.

As I am the one doing these 2.4 patches, I will invariably be asked
whether I intend for the O(1) scheduler to be merged into 2.4. The
answer is a strong NO.

Enjoy,

Robert Love




2002-07-02 22:08:15

by Luigi Genoni

[permalink] [raw]
Subject: Re: [PATCH] O(1) scheduler for 2.4.19-rc1

On 2 Jul 2002, Robert Love wrote:

> Since Ingo recently posted 2.4-ac resyncs, I will refrain.
>
> As I am the one doing these 2.4 patches, I will invariably be asked
> whether I intend for the O(1) scheduler to be merged into 2.4. The
> answer is a strong NO.
>

Of course, I think you know that you will also asked WHY?

Also if I can immagine your reasons, as similar discussions have been
done for preemption patch and so on, and as I said at the times, I Agree.

2.5 is the place for this new and cool stuff.

Luigi

2002-07-02 22:15:35

by Robert Love

[permalink] [raw]
Subject: Re: [PATCH] O(1) scheduler for 2.4.19-rc1

On Tue, 2002-07-02 at 15:10, [email protected] wrote:

> On 2 Jul 2002, Robert Love wrote:
>
> > As I am the one doing these 2.4 patches, I will invariably be asked
> > whether I intend for the O(1) scheduler to be merged into 2.4. The
> > answer is a strong NO.
>
> Of course, I think you know that you will also asked WHY?

Because I do not think 2.4 should be a breeding ground for every new
feature that wets someone's appetite. It should be stable and trusted
before anything else. We also have to worry about architecture
support. Let the scheduler be 2.5's thing.

> Also if I can immagine your reasons, as similar discussions have been
> done for preemption patch and so on, and as I said at the times, I Agree.

I do not think preemption should go in 2.4, either. It too is a 2.5
thing.

> 2.5 is the place for this new and cool stuff.

Agreed.

Robert Love