[openbeos] Re: Off-Topic: Con Kolivas quits Linux. He should know Haiku

On 7/26/07, Salvatore Benedetto <emitrax@xxxxxxxxx> wrote:
> Anyway, talking about scheduling and nice command, instead of having an O(1)
> scheduler, which I'm not a fan of,

Why?

> why not having a scheduler that gets it right 95% of the time,

How about 100% of the time? :)

(yes, that's a teaser. And the fact that 100% is possible is subject
of my next blog post, which I'll publish "when it's done". And no,
it's not possible on Linux, unless they really accept the compromise
of letting go of a very fundamental line that Unices are not really
supposed to cross (Mac OS X excepted).)

> as Con says in his interview, and then give the user
> the possibility to choose which application should get more attention by the
> CPU by simply right clicking on the top bar
> of the window application, instead of having to open a terminal and use
> nice?

This is not intuitive, this is micromanagement of computational
resources. More about this on the eventual blog post.

> Did BeOS have something like that?

Process Controller :)

On 7/26/07, Charlie Clark <charlie@xxxxxxxxxxxxxx> wrote:
> Cogent arguments about scheduling and modularity: let the kernel be
> told a little about it's environment so it knows what to optimise
> for.

Bingo :)

> BeOS did this very well and Linux still sucks at it for some of
> the reasons listed in the article.

The way BeOS did it was a completely wicked hack. And the kernel was
completely ignorant about the surrounding environment; it's the
applications that adapted themselves (app_server, I'm looking at you!)
because they had behind-the-scenes knowledge of how the kernel
behaved.

Hopefully, we won't have to resort to this sort of hacks if I get the
branded semaphore boosting mechanism right.


Cheers,
A.

Other related posts: