From: Ingo Molnar <mingo@kernel.org>
To: Mike Galbraith <efault@gmx.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
Alan Cox <alan@lxorguk.ukuu.org.uk>,
Andi Kleen <andi@firstfloor.org>, Borislav Petkov <bp@alien8.de>,
Nikolay Ulyanitsky <lystor@gmail.com>,
linux-kernel@vger.kernel.org,
Andreas Herrmann <andreas.herrmann3@amd.com>,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Andrew Morton <akpm@linux-foundation.org>,
Thomas Gleixner <tglx@linutronix.de>
Subject: Re: 20% performance drop on PostgreSQL 9.2 from kernel 3.5.3 to 3.6-rc5 on AMD chipsets - bisected
Date: Mon, 17 Sep 2012 12:07:59 +0200 [thread overview]
Message-ID: <20120917100759.GB32463@gmail.com> (raw)
In-Reply-To: <1347869299.6955.156.camel@marge.simpson.net>
* Mike Galbraith <efault@gmx.de> wrote:
> On Sun, 2012-09-16 at 12:57 -0700, Linus Torvalds wrote:
> > On Sat, Sep 15, 2012 at 9:35 PM, Mike Galbraith <efault@gmx.de> wrote:
> > >
> > > Oh, while I'm thinking about it, there's another scenario that could
> > > cause the select_idle_sibling() change to affect pgbench on largeish
> > > packages, but it boils down to preemption odds as well.
> >
> > So here's a possible suggestion..
> >
> > Let's assume that the scheduler code to find the next idle CPU on the
> > package is actually a good idea, and we shouldn't mess with the idea.
>
> We should definitely mess with the idea, as it causes some problems.
>
> > But at the same time, it's clearly an *expensive* idea,
> > which is why you introduced the "only test a single CPU
> > buddy" approach instead. But that didn't work, and you can
> > come up with multiple reasons why it wouldn't work. Plus,
> > quite fundamentally, it's rather understandable that "try to
> > find an idle CPU on the same package" really would be a good
> > idea, right?
>
> I would argue that it did work, it shut down the primary
> source of pain, which I do not believe to be the traversal
> cost, rather the bouncing.
>
> 4 socket 40 core + SMT Westmere box, single 30 sec tbench runs, higher is better:
>
> clients 1 2 4 8 16 32 64 128
> ..........................................................................
> pre 30 41 118 645 3769 6214 12233 14312
> post 299 603 1211 2418 4697 6847 11606 14557
That's a very tempting speedup for a simpler and more
fundamental workload than postgresql's somewhat weird
user-space spinlocks that burn CPU time in user-space
instead of blocking/waiting on a futex.
IIRC mysql does this properly and outperforms postgresql
on this benchmark, in an apples-to-apples configuration?
> 10x at 1 pair shouldn't be traversal, the whole box is
> otherwise idle. We'll do a lot more (ever more futile)
> traversal as load increases, but at the same time, our futile
> attempts fail more frequently, so we shoot ourselves in the
> foot less frequently.
>
> The down side is (appears to be) that I also shut down some
> ~odd case preemption salvation, salvation that only large
> packages will receive.
>
> The problem as I see it is that we're making light tasks _too_
> mobile, turning an optimization into a pessimization for light
> tasks. For longer running tasks this mobility within a large
> package isn't such a big deal, but for fast movers, it hurts a
> lot.
There's not enough time to resolve this for v3.6, so I agree
with the revert - would you be willing to post a v2 of your
original patch? I really think we want your tbench speedups,
quite a few real-world messaging applications use the tbench
patterns of scheduling.
Thanks,
Ingo
next prev parent reply other threads:[~2012-09-17 10:08 UTC|newest]
Thread overview: 115+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-09-14 7:47 20% performance drop on PostgreSQL 9.2 from kernel 3.5.3 to 3.6-rc5 on AMD chipsets Nikolay Ulyanitsky
2012-09-14 18:40 ` Borislav Petkov
2012-09-14 18:51 ` Borislav Petkov
2012-09-14 21:27 ` 20% performance drop on PostgreSQL 9.2 from kernel 3.5.3 to 3.6-rc5 on AMD chipsets - bisected Borislav Petkov
2012-09-14 21:40 ` Peter Zijlstra
2012-09-14 21:44 ` Linus Torvalds
2012-09-14 21:56 ` Peter Zijlstra
2012-09-14 21:59 ` Peter Zijlstra
2012-09-15 3:57 ` Mike Galbraith
2012-09-14 22:01 ` Linus Torvalds
2012-09-14 22:10 ` Peter Zijlstra
2012-09-14 22:20 ` Linus Torvalds
2012-09-14 22:14 ` Borislav Petkov
2012-09-14 21:45 ` Borislav Petkov
2012-09-14 21:42 ` Linus Torvalds
2012-09-15 3:33 ` Mike Galbraith
2012-09-15 16:16 ` Andi Kleen
2012-09-15 16:36 ` Mike Galbraith
2012-09-15 17:08 ` richard -rw- weinberger
2012-09-16 4:48 ` Mike Galbraith
2012-09-15 21:32 ` Alan Cox
2012-09-16 4:35 ` Mike Galbraith
2012-09-16 19:57 ` Linus Torvalds
2012-09-17 8:08 ` Mike Galbraith
2012-09-17 10:07 ` Ingo Molnar [this message]
2012-09-17 10:47 ` Mike Galbraith
2012-09-17 14:39 ` Andi Kleen
2012-09-19 12:35 ` Mike Galbraith
2012-09-19 14:54 ` Ingo Molnar
2012-09-19 15:23 ` Mike Galbraith
2012-09-24 15:00 ` Mel Gorman
2012-09-24 15:23 ` Nikolay Ulyanitsky
2012-09-24 15:53 ` Borislav Petkov
2012-09-24 15:30 ` Peter Zijlstra
2012-09-24 15:51 ` Mike Galbraith
2012-09-24 15:52 ` Linus Torvalds
2012-09-24 16:07 ` Peter Zijlstra
2012-09-24 16:33 ` Linus Torvalds
2012-09-24 16:54 ` Peter Zijlstra
2012-09-25 12:10 ` Hillf Danton
2012-09-24 16:12 ` Peter Zijlstra
2012-09-24 16:30 ` Linus Torvalds
2012-09-24 16:52 ` Borislav Petkov
2012-09-24 16:54 ` Peter Zijlstra
2012-09-24 17:44 ` Peter Zijlstra
2012-09-25 13:23 ` Mel Gorman
2012-09-25 14:36 ` Peter Zijlstra
2012-09-24 18:26 ` Mike Galbraith
2012-09-24 19:12 ` Linus Torvalds
2012-09-24 19:20 ` Borislav Petkov
2012-09-25 1:57 ` Mike Galbraith
2012-09-25 2:11 ` Linus Torvalds
2012-09-25 2:49 ` Mike Galbraith
2012-09-25 3:10 ` Linus Torvalds
2012-09-25 3:20 ` Mike Galbraith
2012-09-25 3:32 ` Linus Torvalds
2012-09-25 3:43 ` Mike Galbraith
2012-09-25 11:58 ` Peter Zijlstra
2012-09-25 13:17 ` Borislav Petkov
2012-09-25 17:00 ` Borislav Petkov
2012-09-25 17:21 ` Linus Torvalds
2012-09-25 18:42 ` Borislav Petkov
2012-09-25 19:08 ` Linus Torvalds
2012-09-26 2:23 ` Mike Galbraith
2012-09-26 17:17 ` Borislav Petkov
2012-09-26 2:00 ` Mike Galbraith
2012-09-26 2:22 ` Linus Torvalds
2012-09-26 2:42 ` Mike Galbraith
2012-09-26 17:15 ` Borislav Petkov
2012-09-26 16:32 ` Borislav Petkov
2012-09-26 18:19 ` Linus Torvalds
2012-09-26 21:37 ` Borislav Petkov
2012-09-27 5:09 ` Mike Galbraith
2012-09-27 5:18 ` Borislav Petkov
2012-09-27 5:44 ` Mike Galbraith
2012-09-27 5:47 ` Ingo Molnar
2012-09-27 5:59 ` Ingo Molnar
2012-09-27 6:34 ` Mike Galbraith
2012-09-27 6:41 ` Ingo Molnar
2012-09-27 6:54 ` Mike Galbraith
2012-09-27 7:10 ` Ingo Molnar
2012-09-27 16:25 ` Borislav Petkov
2012-09-27 17:44 ` Linus Torvalds
2012-09-27 18:05 ` Borislav Petkov
2012-09-27 18:19 ` Linus Torvalds
2012-09-27 18:29 ` Peter Zijlstra
2012-09-27 19:24 ` Borislav Petkov
2012-09-28 3:50 ` Mike Galbraith
2012-09-28 12:30 ` Borislav Petkov
2012-09-27 19:40 ` Linus Torvalds
2012-09-28 4:13 ` Mike Galbraith
2012-09-28 8:37 ` Peter Zijlstra
2012-09-27 7:17 ` david
2012-09-27 7:55 ` Mike Galbraith
2012-09-27 10:20 ` Borislav Petkov
2012-09-27 13:38 ` Mike Galbraith
2012-09-27 16:55 ` david
2012-09-27 4:32 ` Mike Galbraith
2012-09-27 8:21 ` Peter Zijlstra
2012-09-27 16:48 ` david
2012-09-27 17:38 ` Peter Zijlstra
2012-09-27 17:45 ` david
2012-09-27 18:09 ` Peter Zijlstra
2012-09-27 18:15 ` Linus Torvalds
2012-09-27 18:24 ` Borislav Petkov
2012-09-25 1:39 ` Mike Galbraith
2012-09-25 21:11 ` Suresh Siddha
2012-09-25 4:16 ` Mike Galbraith
2012-09-15 4:11 ` Mike Galbraith
[not found] ` <CA+55aFz1A7HbMYS9o-GTS5Zm=Xx8MUD7cR05GMVo--2E34jcgQ@mail.gmail.com>
2012-09-15 4:42 ` Mike Galbraith
2012-09-15 10:44 ` Borislav Petkov
2012-09-15 14:47 ` Mike Galbraith
2012-09-15 15:18 ` Borislav Petkov
2012-09-15 16:13 ` Mike Galbraith
2012-09-15 19:44 ` Borislav Petkov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20120917100759.GB32463@gmail.com \
--to=mingo@kernel.org \
--cc=a.p.zijlstra@chello.nl \
--cc=akpm@linux-foundation.org \
--cc=alan@lxorguk.ukuu.org.uk \
--cc=andi@firstfloor.org \
--cc=andreas.herrmann3@amd.com \
--cc=bp@alien8.de \
--cc=efault@gmx.de \
--cc=linux-kernel@vger.kernel.org \
--cc=lystor@gmail.com \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).