ULE/SCHED_SMP diff for 7.0
Dmitry Morozovsky
marck at rinet.ru
Thu Jul 19 10:00:21 UTC 2007
On Thu, 19 Jul 2007, Jeff Roberson wrote:
JR> > JR> > N Min Max Median Avg
JR> > JR> > x 5 36.33 36.77 36.45 36.486
JR> > JR> > + 5 38.87 39.23 39.03 39.042
JR> >
JR> > Well, I'm a bit puzzled: with new sched_ule.c system+user time slightly
JR> > decreases (within one minute), while real time goes up from 39 to 44
JR> > minutes!
JR> > And most of the time I see 10-50% of idle in top.
JR>
JR> Please try setting sysctl kern.sched.steal_thresh=2 This is the setting I
JR> intend to commit. I also was only testing buildkernel which tends to run
JR> more processes concurrently as compared to buildworld.
Much better! buildworld+buildkernel finished in 37:23, which is almost
identical to SCHED_4BSD. user+sys is stable around 64 mins.
Thanks for your hard work!
Sincerely,
D.Marck [DM5020, MCK-RIPE, DM3-RIPN]
------------------------------------------------------------------------
*** Dmitry Morozovsky --- D.Marck --- Wild Woozle --- marck at rinet.ru ***
------------------------------------------------------------------------
More information about the freebsd-current
mailing list