Candidate multi-console friendly 2.6 scheduling?

[James v. Z. <ja...@dv...>]

Zoltan's patches applayed
> http://www.ltn.lv/~aivils/files/ruby-2.6.10-nofb-20050110.diff.bz2
> 
> Aivils

Now have a working FC3.
Thanks for your work on ruby!

http://kerneltrap.org/node/4493
I saw this and couldn't help but wonder how it might perform,
particularly with the "unusual" load of multiple local consoles.
Towards that end I have applied the zaphod scheduler and gene algorythm
to my 2.6.9 ruby kernel. I have noticed a real change in CPU usage
behaviour. I've observed an adaptec hardware RAID array increase
throughput and other little improvements in behaviour. Reduced idle
load...

http://www.members.westnet.com.au/vanzeeland/Linux/kernel/patch-2.6.9-ruby.vz9x1.diff.gz
Here 'tis. Includes a lot of driver updates and other functionals too.
patch against kernel.org 2.6.9

I am wondering if this scheduler configuration might improve response
for multi console gaming and multi-console loads in general? The task
and anticipatory I/O scheduler tuneable variables will converge towards
the optimum settings for whatever load happens to be running. When the
load changes the schedulers converge towards a new optimum.

I am very curious to hear others thoughts. I've had a couple of days
uptime with it now; doesn't seem to have effected stability at all, and
seeing modest performance improvements. Notably quake3 was very playable
while recompiling a kernel; even on this dual Athlon playability would
previously suffer much more doing this.

Trying two heads of Unreal Tournament 2004 demo, one head in spectator
more chasing the other head around; bear in mind this is a dual 2800+
athlon with 2GbDDR, it took much longer to load the second instance, I
got good agp performance at 1024x768 and PCI at 640x480. The PCI
performance seemed OK, but nothing to write home about, bit choppy.

Quake3 showed improvements with two consoles at 1024x768. Good
performance on both. Previously both had worked ok but there was
noticable moments of lag but playable. Now both are smooth with no laggy
moments and I think I have a higher framerate, at least on the AGP
console.

James

P.S Trying to take it further, I would like to try this gene algorythm
kernel on some single CPU FC3 workstations; I tried to patch the ruby
inlining fixes to 2.6.9, however the following hunk failed, and I know
not how to fix it:
***************
*** 1372,1378 ****
        console_conditional_schedule();
        release_console_sem();

- out:
        return n;
  #undef FLUSH
  }
--- 1372,1377 ----
        console_conditional_schedule();
        release_console_sem();

        return n;
  #undef FLUSH
  }

is there a simple fix?