Message-Id: <200704071650.43905.kernel@kolivas.org>
Date:	Sat, 7 Apr 2007 16:50:43 +1000
From:	Con Kolivas <kernel@...ivas.org>
To:	Ingo Molnar <mingo@...e.hu>
Cc:	Mike Galbraith <efault@....de>,
	linux list <linux-kernel@...r.kernel.org>,
	Andrew Morton <akpm@...ux-foundation.org>,
	ck list <ck@....kolivas.org>
Subject: Re: Ten percent test

On Friday 06 April 2007 20:03, Ingo Molnar wrote:
> * Con Kolivas <kernel@...ivas.org> wrote:
> > > I was more focused on the general case, but all I should have to do
> > > to de-claw all of these sleep exploits is account rr time (only a
> > > couple of lines, done and building now).  It's only a couple of
> > > lines.
> >
> > The more you try to "de-claw" these sleep exploits the less effective
> > you make your precious interactive estimator. Feel free to keep adding
> > endless tweaks to undo the other tweaks in order to try and achieve
> > what SD has by design.
>
> firstly, testing on various workloads Mike's tweaks work pretty well,
> while SD still doesnt handle the high-load case all that well. Note that
> it was you who raised this whole issue to begin with: everything was
> pretty quiet in scheduling interactivity land.

I'm terribly sorry but you have completely missed my intentions then. I was 
_not_ trying to improve mainline's interactivity at all. My desire was to fix 
the unfairness that mainline has, across the board without compromising 
fairness. You said yourself that an approach that fixed a lot and had a small 
number of regressions would be worth it. In a surprisingly ironic turnaround 
two bizarre things happened. People found SD fixed a lot of their 
interactivity corner cases which were showstoppers. That didn't surprise me 
because any unfair design will by its nature get it wrong sometimes. The even 
_more_ surprising thing is that you're now using interactivity as the 
argument against SD. I did not set out to create better interactivity, I set 
out to create widespread fairness without too much compromise to 
interactivity. As I said from the _very first email_, there would be cases of 
interactivity in mainline that performed better.

> (There was one person who 
> reported wide-scale interactivity regressions against mainline but he
> didnt answer my followup posts to trace/debug the scenario.)

That was one user. As I mentioned in an earlier thread, the problem with email 
threads on drawn out issues on lkml is that all that people remember is the 
last one creating noise, and that has only been the noise from Mike for 2 
weeks now. Has everyone forgotten the many many users who reported the 
advantages first up which generated the interest in the first place? Why have 
they stopped reporting? Well the answer is obvious; all the signs suggest 
that SD is slated for mainline. It is on the path, Linus has suggested it and 
now akpm is asking if it's ready for 2.6.22. So they figure there is no point 
testing and replying any further. SD is ready for prime time, finalised and 
does everything I intended it to. This is where I have to reveal to them the 
horrible truth. This is no guarantee it will go in. In fact, this one point 
that you (Ingo) go on and on about is not only a quibble, but you will call 
it an absolute showstopper. As maintainer of the cpu scheduler, in its 
current form you will flatly refuse it goes to mainline citing the 5% of 
cases where interactivity has regressed. So people will tell me to fix it, 
right?... Read on for this to unfold.

> SD has a built-in "interactivity estimator" as well, but hardcoded into
> its design. SD has its own set of ugly-looking tweaks as well - for
> example the prio_matrix.

I'm sorry but this is a mis-representation to me, as I suggested on an earlier 
thread where I disagree about what an interactivity estimator is. The idea of 
fence posts in a clock that are passed as a way of metering out 
earliest-deadline-first in a design is well established. The matrix is simply 
an array designed for O(1) lookups of the fence posts. That is not the same 
as "oh how much have we slept in the last $magic_number period and how much 
extra time should we get for that".

> So it all comes down on 'what interactivity 
> heuristics is enough', and which one is more tweakable. So far i've yet
> to see SD address the hackbench and make -j interactivity
> problems/regression for example, while Mike has been busy addressing the
> 'exploits' reported against mainline.

And BANG there is the bullet you will use against SD from here to eternity. SD 
obeys fairness at all costs. Your interactivity regression is that SD causes 
progressive slowdown with load which by definition is fairness. You 
repeatedly ask me to address it and there is on unfailing truth; the only way 
to address it is to add unfairness to the design. So why don't I? Because the 
simple fact is that any unfairness no matter how carefully administered or 
metered will always have cases where it's wrong. Look at the title of this 
email for example - it's yet another exploit for the mainline sleep/run 
mechanism. This does _not_ mean I'm implying people are logging into servers 
and running ./tenp to hang the machine. What it demonstrates is a way of 
reproducing the scenario which is biting people with real world loads. It's 
entirely believable that a simple p2p app could be behaving like tenp, only 
generating a small load and it could take ages to log in and use the console. 
Willy has complained this is why people stick to 2.4. Sure I can create 
interactivity tweaks worse than anyone else. I will not, though, because that 
precisely undoes what is special about SD. It never looks backwards, and is 
predictable to absurdity. So you'll argue that mainline can manage it 
below...

> > You'll end up with an incresingly complex state machine design of
> > interactivity tweaks and interactivity throttlers all fighting each
> > other to the point where the intearactivity estimator doesn't do
> > anything. [...]
>
> It comes down to defining interactivity by scheduling behavior, and
> making that definition flexible. SD's definition of interactivity is
> rigid (but it's still behavior-based, so not fundamentally different
> from an explicit 'interactivity estimator'), and currently it does not
> work well under high load. But ... i'm still entertaining the notion
> that it might be good enough, but you've got to demonstrate the design's
> flexibility.

I have yet to see someone find an "exploit" for SD's current design. Mainline 
is all about continually patching up the intrinsic design (and fixing this 
one test case is not the be all and end all).

> furthermore, your description does not match my experience when using
> Mike's tweaks and comparing it to SD on the same hardware. According to
> your claim i should have seen regressions popping up in various,
> already-fixed corners, but it didnt happen in practice. But ... i'm
> awaiting further SD and Mike tweaks, the race certainly looks
> interesting ;)

Well you see a race. I do not. I see a flat predictable performance from SD 
where there will always be slowdown with load. I have no intention of 
changing that. Mike is making an admirable attempt to fix issues as they are 
pointed out. You say there are no regressions but I see absolutely no testers 
of his patches besides himself and you. If I introduce any unfairness based 
on sleep behaviour into SD I'll be undoing the whole point of the design and 
end up chasing new regressions. So I won't quibble over the numbers. SD has 
produced a lot of improvements and fairness that mainline struggles with ever 
increasing patches to emulate, but SD does so at the expense of proportional 
slowdown with load. At least I accept that and will no longer put my health 
at risk trying to "fix" it by "breaking" it. SD is done.

I feel sorry for the many users out there who are simply "waiting for it to 
end up in mainline" who just discovered you will veto it on that basis. 
lwn.net had it wrong; this was far more painful than any previous attempt to 
get anything into mainline.

My health has been so badly affected by this that I've been given an ultimatum 
and must turn my computer off till I get well now which may be weeks. I 
already know the massive flameage and last-word comments that are likely to 
be fired off before the inevitable decision to veto it.

> 	Ingo

さようなら

-- 
-ck
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists