Subject: Re: sucky performance on i386/1.3I
To: None <,>
From: Ross Harvey <>
List: tech-kern
Date: 01/23/1999 01:38:10
Regarding the general problem, the one very effectively demonstrated by
Jens Nilsson's program posted by ragge, I think I've fixed that. (And thanks
to Charles Hannum for confirming the analysis.)

You can get the mod by updating your /sys/kern/kern_exit.c module.  For
now, you can turn the it off and back on by patching `chargeparent' to 0
and (back to) 1.

I believe this problem has always been there, but as it is a scheduler
effect, perhaps we have recently aggravated it. For example, a speedup of
some kind in fork(2) or exit(2) or a slowdown in X might have tipped a
script below the X server in relative CPU usage, triggering this syndrome.
Examples of things that might have sped us up are vfork(2) and UVM. While
the reaper is overhead, it does hide CPU use a bit and might also be a

Regarding Roland's observation about nice(2) problems on alpha: urk, that's
a known and analyzed also affects i386 somewhat, though not nearly
as badly. While we are banging on the scheduler I'll see about fixing that.