Re: computer becomes slow when compiling something

On Sun, Aug 26, 2007 at 09:35:56PM -0700, Kevin Oberman wrote:
> > Date: Mon, 27 Aug 2007 12:25:38 +0800
> > From: Ganbold <ganbold_at_micom.mng.net>
> > Sender: owner-freebsd-current_at_freebsd.org
> > 
> > Hi,
> > 
> > I'm running FreeBSD 7.0-CURRENT with SCHED_ULE, INVARIANTS, WITNESS 
> > enabled kernel.
> > 
> > daemon# uname -an
> > FreeBSD daemon.micom.mng.net 7.0-CURRENT FreeBSD 7.0-CURRENT #2: Thu Aug 
> > 23 17:59:17 ULAT 2007     
> > tsgan_at_daemon.micom.mng.net:/usr/obj/usr/src/sys/GDAEMON  i386
> > daemon#
> > 
> > When compiling something (buildworld or making wine for example) inside 
> > from X/gnome
> > my computer becomes very slow.
> > 
> > top shows while compiling wine:
> > 
> > last pid: 38660;  load averages:  3.10,  2.24,  
> > 1.33                                                                                 
> > up 3+02:07:05  12:11:38
> > 106 processes: 3 running, 102 sleeping, 1 zombie
> > CPU states: 90.2% user,  0.0% nice,  9.8% system,  0.0% interrupt,  0.0% 
> > idle
> > Mem: 704M Active, 63M Inact, 168M Wired, 39M Cache, 111M Buf, 21M Free
> > Swap: 2048M Total, 195M Used, 1853M Free, 9% Inuse
> > 
> >   PID USERNAME    THR PRI NICE   SIZE    RES STATE    TIME   WCPU COMMAND
> >   902 tsgan         1  96    0   147M   134M RUN    929:41  3.96% Xorg
> > 38659 root          1  96    0 15556K 12256K RUN      0:00  3.47% cc1
> >  1206 tsgan         1  44   19   156M 36940K select 178:06  0.00% 
> > operapluginwrapper
> >   978 tsgan         1  44    0   212M   173M select 103:26  0.00% opera
> >   954 tsgan         1  44    0 35236K 18372K select  11:59  0.00% 
> > wnck-applet
> >   975 tsgan         7  44    0   198M   159M ucond    8:53  0.00% 
> > thunderbird-bin
> >   922 tsgan         1  44    0  4808K  1460K select   4:43  0.00% gam_server
> >   956 tsgan         1  44    0 35324K 16328K select   4:36  0.00% 
> > mixer_applet2
> >   962 tsgan         1  44    0 13736K  5936K select   1:57  0.00% 
> > gnome-screensaver
> >  1224 tsgan         1  44    0 70232K 28900K select   1:33  0.00% pidgin
> >  1591 tsgan         1  44   19  5520K   772K select   1:17  0.00% 
> > operapluginwrapper
> >   772 root          1  44    0  4496K   928K select   1:03  0.00% 
> > hald-addon-storage
> >   930 tsgan         1  44    0 16204K  8744K select   0:48  0.00% metacity
> >   441 root          1  44    0  3276K   592K select   0:45  0.00% moused
> >   765 haldaemon     1  44    0  6164K  2252K select   0:41  0.00% hald
> > ...
> > 
> > Is it due to SCHED_ULE makes a process CPU greedy and that is why my 
> > computer becomes slow?
> > Or it is something else? What SCHED_ULE sysctl knobs should I test here?
> > As I recall correctly I have never experienced such problems until 
> > recently.
> > Maybe I'm wrong here.
> 
> I've seen it for a while. You are running out of RAM and I have found
> that once the swap use starts growing, things get VERY slow. I still
> think something is wrong, but memory is cheap and an extra 1/2 Gig can
> make a huge difference. I have not seen any difference between
> schedulers...about the same with 4BSD and ULE.

No, this happens on the machine with a plenty of free RAM too, for instance

Mem: 187M Active, 868M Inact, 162M Wired, 20M Cache, 112M Buf, 764M Free
Swap: 8192M Total, 8192M Free

SCHED_4BSD was almost unusable for me, SCHED_ULE gives somewhat better
interactivity.

It seems that fork() gives too high priority for the new process.