From: Linus Torvalds <torvalds@linux-foundation.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>,
Minchan Kim <minchan.kim@gmail.com>,
"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
cl@linux-foundation.org,
"hugh.dickins" <hugh.dickins@tiscali.co.uk>,
Nick Piggin <nickpiggin@yahoo.com.au>,
Ingo Molnar <mingo@elte.hu>
Subject: Re: [RFC][PATCH 6/8] mm: handle_speculative_fault()
Date: Fri, 8 Jan 2010 09:22:14 -0800 (PST) [thread overview]
Message-ID: <alpine.LFD.2.00.1001080911340.7821@localhost.localdomain> (raw)
In-Reply-To: <1262969610.4244.36.camel@laptop>
On Fri, 8 Jan 2010, Peter Zijlstra wrote:
> On Tue, 2010-01-05 at 20:20 -0800, Linus Torvalds wrote:
> >
> > Yeah, I should have looked more at your callchain. That's nasty. Much
> > worse than the per-mm lock. I thought the page buffering would avoid the
> > zone lock becoming a huge problem, but clearly not in this case.
>
> Right, so I ran some numbers on a multi-socket (2) machine as well:
>
> pf/min
>
> -tip 56398626
> -tip + xadd 174753190
> -tip + speculative 189274319
> -tip + xadd + speculative 200174641
>
> [ variance is around 0.5% for this workload, ran most of these numbers
> with --repeat 5 ]
That's a huge jump. It's clear that the spinlock-based rwsem's simply
suck. The speculation gets rid of some additional mmap_sem contention,
but at least for two sockets it looks like the rwsem implementation was
the biggest problem by far.
> At both the xadd/speculative point the workload is dominated by the
> zone->lock, the xadd+speculative removes some of the contention, and
> removing the various RSS counters could yield another few percent
> according to the profiles, but then we're pretty much there.
I don't know if worrying about a few percent is worth it. "Perfect is the
enemy of good", and the workload is pretty dang artificial with the whole
"remove pages and re-fault them as fast as you can".
So the benchmark is pointless and extreme, and I think it's not worth
worrying too much about details. Especially when compared to just the
*three-fold* jump from just the fairly trivial rwsem implementation change
(with speculation on top of it then adding another 15% improvement -
nothing to sneeze at, but it's still in a different class).
Of course, larger numbers of sockets will likely change the situation, but
at the same time I do suspect that workloads designed for hundreds of
cores will need to try to behave better than that benchmark anyway ;)
> One way around those RSS counters is to track it per task, a quick grep
> shows its only the oom-killer and proc that use them.
>
> A quick hack removing them gets us: 203158058
Yeah, well.. After that 200% and 15% improvement, a 1.5% improvement on a
totally artificial benchmark looks less interesting.
Because let's face it - if your workload does several million page faults
per second, you're just doing something fundamentally _wrong_.
Linus
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2010-01-08 17:23 UTC|newest]
Thread overview: 121+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-01-04 18:24 [RFC][PATCH 0/8] Speculative pagefault -v3 Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 1/8] mm: Remove pte reference from fault path Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 2/8] mm: Speculative pagefault infrastructure Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 3/8] mm: Add vma sequence count Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 4/8] mm: RCU free vmas Peter Zijlstra
2010-01-05 2:43 ` Paul E. McKenney
2010-01-05 8:28 ` Peter Zijlstra
2010-01-05 16:05 ` Paul E. McKenney
2010-01-04 18:24 ` [RFC][PATCH 5/8] mm: Speculative pte_map_lock() Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 6/8] mm: handle_speculative_fault() Peter Zijlstra
2010-01-05 0:25 ` KAMEZAWA Hiroyuki
2010-01-05 3:13 ` Linus Torvalds
2010-01-05 8:17 ` Peter Zijlstra
2010-01-05 8:57 ` Peter Zijlstra
2010-01-05 15:34 ` Linus Torvalds
2010-01-05 15:40 ` Al Viro
2010-01-05 16:10 ` Linus Torvalds
2010-01-06 15:41 ` Peter Zijlstra
2010-01-05 9:37 ` Peter Zijlstra
2010-01-05 23:35 ` Linus Torvalds
2010-01-05 4:29 ` Minchan Kim
2010-01-05 4:43 ` KAMEZAWA Hiroyuki
2010-01-05 5:10 ` Linus Torvalds
2010-01-05 5:30 ` KAMEZAWA Hiroyuki
2010-01-05 7:39 ` KAMEZAWA Hiroyuki
2010-01-05 15:26 ` Linus Torvalds
2010-01-05 16:14 ` Linus Torvalds
2010-01-05 17:25 ` Andi Kleen
2010-01-05 17:47 ` Christoph Lameter
2010-01-05 18:00 ` Andi Kleen
2010-01-05 17:55 ` Linus Torvalds
2010-01-05 18:13 ` Christoph Lameter
2010-01-05 18:25 ` Linus Torvalds
2010-01-05 18:46 ` Christoph Lameter
2010-01-05 18:56 ` Linus Torvalds
2010-01-05 19:15 ` Christoph Lameter
2010-01-05 19:28 ` Linus Torvalds
2010-01-05 18:55 ` Paul E. McKenney
2010-01-05 19:08 ` Linus Torvalds
2010-01-05 19:23 ` Paul E. McKenney
2010-01-05 20:29 ` Peter Zijlstra
2010-01-05 20:46 ` Linus Torvalds
2010-01-05 21:00 ` Linus Torvalds
2010-01-05 23:29 ` Paul E. McKenney
2010-01-06 0:22 ` KAMEZAWA Hiroyuki
2010-01-06 1:37 ` Linus Torvalds
2010-01-06 2:52 ` KAMEZAWA Hiroyuki
2010-01-06 3:27 ` Linus Torvalds
2010-01-06 3:56 ` KAMEZAWA Hiroyuki
2010-01-06 4:20 ` Linus Torvalds
2010-01-06 7:06 ` KAMEZAWA Hiroyuki
2010-01-06 7:49 ` Minchan Kim
2010-01-06 9:39 ` Linus Torvalds
2010-01-07 1:00 ` KAMEZAWA Hiroyuki
2010-01-08 16:53 ` Peter Zijlstra
2010-01-08 17:22 ` Linus Torvalds [this message]
2010-01-08 17:43 ` Christoph Lameter
2010-01-08 17:52 ` Linus Torvalds
2010-01-08 18:33 ` Christoph Lameter
2010-01-08 18:46 ` Andi Kleen
2010-01-08 18:56 ` Christoph Lameter
2010-01-08 19:10 ` Andi Kleen
2010-01-08 19:11 ` Linus Torvalds
2010-01-08 19:28 ` Andi Kleen
2010-01-08 19:39 ` Linus Torvalds
2010-01-08 19:42 ` Linus Torvalds
2010-01-08 21:36 ` Linus Torvalds
2010-01-08 21:46 ` Christoph Lameter
2010-01-08 22:43 ` Linus Torvalds
2010-01-08 22:43 ` Linus Torvalds
2010-01-09 14:47 ` Ed Tomlinson
2010-01-10 5:27 ` Nitin Gupta
2010-01-05 15:14 ` Christoph Lameter
2010-01-05 8:18 ` Peter Zijlstra
2010-01-05 6:00 ` Minchan Kim
2010-01-05 4:48 ` Linus Torvalds
2010-01-05 6:09 ` Minchan Kim
2010-01-05 6:09 ` KAMEZAWA Hiroyuki
2010-01-05 6:24 ` Minchan Kim
2010-01-05 8:35 ` Peter Zijlstra
2010-01-05 13:45 ` Arjan van de Ven
2010-01-05 14:15 ` Andi Kleen
2010-01-05 15:17 ` Christoph Lameter
2010-01-06 3:22 ` Arjan van de Ven
2010-01-07 16:11 ` Christoph Lameter
2010-01-07 16:19 ` Linus Torvalds
2010-01-07 16:31 ` Linus Torvalds
2010-01-07 16:34 ` Paul E. McKenney
2010-01-07 16:36 ` Christoph Lameter
2010-01-08 4:49 ` Arjan van de Ven
2010-01-08 5:00 ` Linus Torvalds
2010-01-08 15:51 ` Christoph Lameter
2010-01-09 15:55 ` Arjan van de Ven
2010-01-07 17:22 ` Peter Zijlstra
2010-01-07 17:36 ` Linus Torvalds
2010-01-07 17:49 ` Linus Torvalds
2010-01-07 18:00 ` Peter Zijlstra
2010-01-07 18:15 ` Linus Torvalds
2010-01-07 21:49 ` Peter Zijlstra
2010-01-07 18:44 ` Linus Torvalds
2010-01-07 19:20 ` Paul E. McKenney
2010-01-07 20:06 ` Linus Torvalds
2010-01-07 20:25 ` Paul E. McKenney
2010-01-07 19:24 ` Christoph Lameter
2010-01-07 20:08 ` Linus Torvalds
2010-01-07 20:13 ` Linus Torvalds
2010-01-07 21:44 ` Peter Zijlstra
2010-01-07 22:33 ` Linus Torvalds
2010-01-08 0:23 ` KAMEZAWA Hiroyuki
2010-01-08 0:25 ` KAMEZAWA Hiroyuki
2010-01-08 0:39 ` Linus Torvalds
2010-01-08 0:41 ` Linus Torvalds
2010-01-07 23:51 ` Rik van Riel
2010-01-04 18:24 ` [RFC][PATCH 7/8] mm,x86: speculative pagefault support Peter Zijlstra
2010-01-04 18:24 ` [RFC][PATCH 8/8] mm: Optimize pte_map_lock() Peter Zijlstra
2010-01-04 21:41 ` [RFC][PATCH 0/8] Speculative pagefault -v3 Rik van Riel
2010-01-04 21:46 ` Peter Zijlstra
2010-01-04 23:20 ` Rik van Riel
2010-01-04 21:59 ` Christoph Lameter
2010-01-05 0:28 ` KAMEZAWA Hiroyuki
2010-01-05 2:26 ` Minchan Kim
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=alpine.LFD.2.00.1001080911340.7821@localhost.localdomain \
--to=torvalds@linux-foundation.org \
--cc=cl@linux-foundation.org \
--cc=hugh.dickins@tiscali.co.uk \
--cc=kamezawa.hiroyu@jp.fujitsu.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=minchan.kim@gmail.com \
--cc=mingo@elte.hu \
--cc=nickpiggin@yahoo.com.au \
--cc=paulmck@linux.vnet.ibm.com \
--cc=peterz@infradead.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).