Re: TREE_SRCU slows hotplug by factor ~16

[Paul E. McKenney]

On Wed, Apr 26, 2017 at 05:49:59PM +0200, Mike Galbraith wrote:
> On Wed, 2017-04-26 at 08:44 -0700, Paul E. McKenney wrote:
> > On Wed, Apr 26, 2017 at 05:26:20PM +0200, Mike Galbraith wrote:
> > > On Wed, 2017-04-26 at 07:31 -0700, Paul E. McKenney wrote:
> > > 
> > > > And a sneak preview, semi-tested.  If you get a chance to run this, please
> > > > let me know now it goes.
> > > 
> > > That took 'time stress-cpu-hotplug.sh' down to 48s, close to classic.
> > 
> > Woo-hoo!!!  ;-)
> > 
> > And thank you for your testing efforts!
> > 
> > Should I be comparing this with the 55s number from your initial email,
> > or to the 39s number?
> 
> Should be the 39, but I'll get you a new tree vs classic number in this
> same tree after I finish bisecting TSC going wonky.

Sounds good!!!

In theory, given 4-millisecond SRCU grace periods and 225,000 of them
per run (estimated based on experience and on numbers you have already
provided me), the 0x3ff should be worth only about one second of delay
over the full test.  On a really good day, I could probably invent some
statistics to estimate the slowdown from the 50-microsecond cutoff, but
today is not quite that good a day.  And besides, the only difference
between theory and practice is that in theory they are both the same.  ;-)

So I look forward to seeing your measurements!  And best of everything
with the wonky TSC!!!

							Thanx, Paul