Re: UDP-U stream performance regression on 32-rc1 kernel

[Mike Galbraith <efault@gmx.de>]

On Tue, 2009-11-03 at 12:33 +0800, Zhang, Yanmin wrote:
> On Tue, 2009-11-03 at 11:47 +0800, Alex Shi wrote:
> > We found the UDP-U 1k/4k stream of netperf benchmark have some
> > performance regression from 10% to 20% on our Tulsa and some NHM
> > machines. 
> 
> perf events shows function find_busiest_group consumes about 4.5% cpu time
> with the patch while it only consumes 0.5% cpu time without the patch.
> 
> The communication between netperf client and netserver is very fast.
> When netserver receives a message and there is no new message available,
> it goes to sleep and scheduler calls idle_balance => load_balance_newidle.
> load_balance_newidle spends too much time and a new message arrives quickly
> before load_balance_newidle ends.

I have a similar problem wrt ramp-up and affinity, so will certainly be
doing battle with the thing here.

It's harming mysql+oltp and pgsql+oltp ramp up, and modest load in
general by pulling at the first micro-sleep.  After twiddling
wake_affine() to spread to a shared cache, newidle comes along and
throws a wrench into my plans an eyeblink later.

> As the comments in the patch say hackbench benefits from it, I tested hackbench
> on Nehalem and core2 machines. hackbench does benefit from it, about 6% on
> nehalem machines, but doesn't benefit on core2 machines.

It depends a lot on the load.  I have a testcase which spawns threads at
a ~high rate.  There, turning it off costs ~42% on my little Q6600 box.
It's also a modest utilization win for a kbuild.

In any case though, it certainly wants a couple fangs pulled.

	-Mike