On 2007-04-10 at 03:15:55 [+0200], André Braga <meianoite@xxxxxxxxx> wrote:
> On 4/9/07, Ingo Weinhold <bonefish@xxxxxxxxxxxxxxx> wrote:
>
> > please repost on the kernel mailing list. I believe you miss some kernel
> > developers (e.g. Travis?) when posting here.
>
> I did!
>
>
http://sourceforge.net/mailarchive/forum.php?thread_name=2ad73a0704081600k6eeb8a5bh6a3231f67e588725%40mail.gmail.com&forum_name=open-beos-kernel-devel
>
> Perhaps some spam filter thought my mail was junk? =P
Ah, no, sorry. I just seemed to have missed it there.
> P.s.: no remarks, no comments, no suggestions whatsoever?
I'm not exactly knowledgeable in the scheduling area. I last heard
something about it as a student 7 or 8 years ago, and more intriguing
algorithms weren't covered either.
Generally, we're focussing on desktop use, not server (max throughput), nor
real time applications. So, I believe, we want a scheduler with low
latencies and acceptable throughput. Performance and complexity should be
reasonably balanced. IMHO, a GA approach is over the top, as is your
specialized memory allocator idea.
Regarding memory usage, I think it's not that much of an issue. While it's
generally prudent to have an eye on kernel memory usage, a few bytes more
in struct thread won't be a problem -- even with thousand threads (a rather
unusual case I'd say) it's only some KB after all. So, freely add what you
need and don't do nasty tricks just to save a few bytes.
CU, Ingo
