Re: [PATCH 0/2] RFC: readd fair sleepers for server systems

[Mike Galbraith <efault@gmx.de>]

On Tue, 2012-05-22 at 09:11 +0200, Christian Ehrhardt wrote: 
> 
> On 05/21/2012 08:17 PM, Mike Galbraith wrote:
> > On Mon, 2012-05-21 at 17:45 +0200, Martin Schwidefsky wrote:
> >> our performance team found a performance degradation with a recent
> >> distribution update in regard to fair sleepers (or the lack of fair
> >> sleepers). On s390 we used to run with fair sleepers disabled.
> >>
> >> We see the performance degradation with our network benchmark and fair
> >> sleepers enabled, the largest hit is on virtual connections:
> >
> > I can see you wanting the feature back.  You guys apparently do not
> > generally run mixed loads on your boxen, else you wouldn't want to turn
> > the scheduler into a tick granularity scheduler, but why compile time?
> > If the fast path branch isn't important, and given it only became
> > important while I was trying to scrape a few cycles together, why not
> > just restore the feature as it used to exist under the pretext that you
> > need it, and others may as well, so we eat the branch in the interest of
> > general flexibility, and call removal a booboo?
> >
> > -Mike
> >
> 
> If "eating the branches" is fine for everyone s390 can surely live with 
> it. The intention to make it configurable, was to allow systems that 
> really never want it, to be still able to avoid the branch.
> 
> By that everyone can configure it the way they want it and we avoid 
> another modification of the same code over and over again.

Ok.  Features have become cheaper, but we can still use every cycle we
can get our grubby mitts on.

-Mike