Re: Periodic rant about SCHED_ULE

From: Shaun Loo <>
Date: Tue, 28 Mar 2023 10:34:33 UTC

On Sun, Mar 26, 2023 at 1:24 PM George Mitchell < 
<>> wrote:

    On 3/25/23 23:42, Peter wrote:
     > Quoting George Mitchell <
     > On 3/25/23 11:47, Peter wrote:
     >>> You're welcome. Can I get a success/failure report?
     >>> [...]
     >> Thanks for your help.  Regretfully, I have to report that the patch
     >> did not really help in my case (make buildworld while running
     >> But I have belatedly realized that it's a really good idea to use
     >> make -j12 to build kernels and the world.  With 4BSD and dnetc
     >> running, I can now buildworld in 3512 seconds (down from 20477), and
     >> with ULE in 14193 seconds (down from 50290).  Peter's patch got it
     >> down to 13755 seconds.
     > Thank You for testing. So at least it doesn't make things worse.
     > Back then 5 years ago when I made that, I think I've seen a few
     > more strange things in that code, but they didn't bother me
     > at that point. Probably there is more work to do there...

    To oversimplify the discussion I've seen so far on this thread:

    It would be really nice if the schedulers were kernel loadable modules.
    GSoC project?

I'm definitely eager to put forth a GSoC project to make schedulers into
kernel modules! I'll be reviewing how to write kernel modules such that I'll
be in a good position when GSoC comes. I've been checking this thread
on-and off for the past few days so I definitely think I'm in a good 
to work on exactly this!

Is anyone in this thread willing and able to be my mentor? I'll try to 
come up
with a compelling proposal in the next couple days or so to get an 
in right before the April 4th deadline.


    Whether SCHED_ULE or SCHED_4BSD should be the default scheduler in the
    GENERIC kernel is a contentious discussion, but perhaps we need to have
    that discussion.                                           -- George