From: Waiman Long <waiman.long@hp.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>,
Benjamin Herrenschmidt <benh@kernel.crashing.org>,
Alexander Viro <viro@zeniv.linux.org.uk>,
Jeff Layton <jlayton@redhat.com>,
Miklos Szeredi <mszeredi@suse.cz>, Ingo Molnar <mingo@redhat.com>,
Thomas Gleixner <tglx@linutronix.de>,
linux-fsdevel <linux-fsdevel@vger.kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
Peter Zijlstra <peterz@infradead.org>,
Steven Rostedt <rostedt@goodmis.org>,
Andi Kleen <andi@firstfloor.org>,
"Chandramouleeswaran, Aswin" <aswin@hp.com>,
"Norton, Scott J" <scott.norton@hp.com>
Subject: Re: [PATCH v7 1/4] spinlock: A new lockref structure for lockless update of refcount
Date: Fri, 30 Aug 2013 16:15:13 -0400 [thread overview]
Message-ID: <5220FD51.2010709@hp.com> (raw)
In-Reply-To: <CA+55aFzqJ5So9Hj1H0C86e4s13FuEEC_n6ubgJ6SexEDB5rYHg@mail.gmail.com>
On 08/30/2013 03:33 PM, Linus Torvalds wrote:
> On Fri, Aug 30, 2013 at 12:20 PM, Waiman Long<waiman.long@hp.com> wrote:
>> Below is the perf data of my short workloads run in an 80-core DL980:
> Ok, that doesn't look much like d_lock any more. Sure, there's a small
> amount of spinlocking going on with lockref being involved, but on the
> whole even that looks more like getcwd and other random things.
Yes, d_lock contention isn't a major one in the perf profile. However,
sometimes a small improvement can lead to a noticeable improvement in
performance.
> I do agree that getcwd() can probably be hugely optimized. Nobody has
> ever bothered, because it's never really performance-critical, and I
> think AIM7 ends up just doing something really odd. I bet we could fix
> it entirely if we cared enough.
The prepend_path() isn't all due to getcwd. The correct profile should be
|--12.81%-- prepend_path
| |
| |--67.35%-- d_path
| | |
| | |--60.72%--
proc_pid_readlink
| | | sys_readlinkat
| | | sys_readlink
| | |
system_call_fastpath
| | | __GI___readlink
| | | 0x302f64662f666c
| | |
| | --39.28%--
perf_event_mmap_event
| |
| --32.65%-- sys_getcwd
| system_call_fastpath
| __getcwd
Yes, the perf subsystem itself can contribute a sizeable portion of the
spinlock contention. In fact, I have also applied my seqlock patch that
was sent a while ago to the test kernel in order to get a more accurate
perf profile. The seqlock patch will allow concurrent d_path() calls
without one blocking the others. In the 240-core prototype machine, it
was not possible to get an accurate perf profile for some workloads
because more than 50% of the time was spent in spinlock contention due
to the use of perf. An accurate perf profile can only be obtained in
those cases by applying my lockref and seqlock patches. I hope someone
will have the time to review my seqlock patch to see what additional
changes will be needed. I really like to see it merged in some form to
3.12.
> I just wonder if it's even worth it (I assume AIM7 is something HP
> uses internally, because I've never really heard of anybody else
> caring)
Our performance group is actually pretty new. It was formed 2 years ago
and we began actively participating in the Linux kernel development just
in the past year.
We use the AIM7 benchmark internally primarily because it is easy to run
and cover quite a lot of different areas in the kernel. We are also
using specJBB and SwingBench for performance benchmarking problem. We
are also trying to look for more benchmarks to use in the future.
Regards,
Longman
next prev parent reply other threads:[~2013-08-30 20:15 UTC|newest]
Thread overview: 151+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-08-06 3:12 [PATCH v7 0/4] Lockless update of reference count protected by spinlock Waiman Long
2013-08-06 3:12 ` [PATCH v7 1/4] spinlock: A new lockref structure for lockless update of refcount Waiman Long
2013-08-29 1:40 ` Linus Torvalds
2013-08-29 4:44 ` Benjamin Herrenschmidt
2013-08-29 7:00 ` Ingo Molnar
2013-08-29 16:43 ` Linus Torvalds
2013-08-29 19:25 ` Linus Torvalds
2013-08-29 23:42 ` Linus Torvalds
2013-08-30 0:26 ` Benjamin Herrenschmidt
2013-08-30 0:49 ` Linus Torvalds
2013-08-30 2:06 ` Michael Neuling
2013-08-30 2:30 ` Benjamin Herrenschmidt
2013-08-30 2:35 ` Linus Torvalds
2013-08-30 2:45 ` Benjamin Herrenschmidt
2013-08-30 2:31 ` Linus Torvalds
2013-08-30 2:43 ` Benjamin Herrenschmidt
2013-08-30 7:16 ` Ingo Molnar
2013-08-30 15:28 ` Linus Torvalds
2013-08-30 3:12 ` Waiman Long
2013-08-30 3:54 ` Linus Torvalds
2013-08-30 7:55 ` Sedat Dilek
2013-08-30 8:10 ` Sedat Dilek
2013-08-30 9:27 ` Sedat Dilek
2013-08-30 9:48 ` Ingo Molnar
2013-08-30 9:56 ` Sedat Dilek
2013-08-30 9:58 ` Sedat Dilek
2013-08-30 10:29 ` Sedat Dilek
2013-08-30 10:36 ` Peter Zijlstra
2013-08-30 10:44 ` Sedat Dilek
2013-08-30 10:46 ` Sedat Dilek
2013-08-30 10:52 ` Peter Zijlstra
2013-08-30 10:57 ` Sedat Dilek
2013-08-30 14:05 ` Sedat Dilek
2013-08-30 11:19 ` Sedat Dilek
2013-08-30 10:38 ` Sedat Dilek
2013-08-30 15:34 ` Linus Torvalds
2013-08-30 15:38 ` Sedat Dilek
2013-08-30 16:12 ` Steven Rostedt
2013-08-30 16:16 ` Sedat Dilek
2013-08-30 18:42 ` Linus Torvalds
2013-08-30 16:32 ` Linus Torvalds
2013-08-30 16:37 ` Sedat Dilek
2013-08-30 16:52 ` Linus Torvalds
2013-08-30 17:11 ` Sedat Dilek
2013-08-30 17:26 ` Linus Torvalds
2013-09-01 10:01 ` Sedat Dilek
2013-09-01 10:33 ` Sedat Dilek
2013-09-01 15:32 ` Linus Torvalds
2013-09-01 15:45 ` Sedat Dilek
2013-09-01 15:55 ` Linus Torvalds
2013-09-02 10:30 ` Sedat Dilek
2013-09-02 16:09 ` David Ahern
2013-09-01 20:59 ` Linus Torvalds
2013-09-01 21:23 ` Al Viro
2013-09-01 22:16 ` Linus Torvalds
2013-09-01 22:35 ` Al Viro
2013-09-01 22:44 ` Al Viro
2013-09-01 22:58 ` Linus Torvalds
2013-09-01 22:48 ` Linus Torvalds
2013-09-01 23:30 ` Al Viro
2013-09-02 0:12 ` Linus Torvalds
2013-09-02 0:50 ` Linus Torvalds
2013-09-02 7:05 ` Ingo Molnar
2013-09-02 16:44 ` Linus Torvalds
2013-09-03 10:15 ` Ingo Molnar
2013-09-03 15:41 ` Linus Torvalds
2013-09-03 18:34 ` Linus Torvalds
2013-09-03 19:19 ` Ingo Molnar
2013-09-03 21:05 ` Linus Torvalds
2013-09-03 21:13 ` Linus Torvalds
2013-09-03 21:34 ` Linus Torvalds
2013-09-03 21:39 ` Linus Torvalds
2013-09-03 14:08 ` Pavel Machek
2013-09-03 22:37 ` Sedat Dilek
2013-09-03 22:55 ` Dave Jones
2013-09-03 23:05 ` Sedat Dilek
2013-09-03 23:15 ` Dave Jones
2013-09-03 23:20 ` Sedat Dilek
2013-09-03 23:45 ` Sedat Dilek
2013-08-30 18:33 ` Waiman Long
2013-08-30 18:53 ` Linus Torvalds
2013-08-30 19:20 ` Waiman Long
2013-08-30 19:33 ` Linus Torvalds
2013-08-30 20:15 ` Waiman Long [this message]
2013-08-30 20:43 ` Linus Torvalds
2013-08-30 20:54 ` Al Viro
2013-08-30 21:03 ` Linus Torvalds
2013-08-30 21:44 ` Al Viro
2013-08-30 22:30 ` Linus Torvalds
2013-08-31 21:23 ` Al Viro
2013-08-31 22:49 ` Linus Torvalds
2013-08-31 23:27 ` Al Viro
2013-09-01 0:13 ` Al Viro
2013-09-01 17:48 ` Al Viro
2013-09-09 8:30 ` Peter Zijlstra
2013-08-30 21:10 ` Waiman Long
2013-08-30 21:22 ` Linus Torvalds
2013-08-30 21:30 ` Al Viro
2013-08-30 21:42 ` Waiman Long
2013-08-30 19:40 ` Al Viro
2013-08-30 19:52 ` Waiman Long
2013-08-30 20:26 ` Al Viro
2013-08-30 20:35 ` Waiman Long
2013-08-30 20:48 ` Al Viro
2013-08-31 2:02 ` Waiman Long
2013-08-31 2:35 ` Al Viro
2013-08-31 2:42 ` Al Viro
2013-09-02 19:25 ` Waiman Long
2013-09-03 6:01 ` Ingo Molnar
2013-09-03 7:24 ` Sedat Dilek
2013-09-03 15:38 ` Linus Torvalds
2013-09-03 15:14 ` Waiman Long
2013-09-03 15:34 ` Linus Torvalds
2013-09-03 19:09 ` Linus Torvalds
2013-09-03 21:01 ` Waiman Long
2013-09-04 14:52 ` Waiman Long
2013-09-04 15:14 ` Linus Torvalds
2013-09-04 19:25 ` Waiman Long
2013-09-04 21:34 ` Linus Torvalds
2013-09-05 2:35 ` Waiman Long
2013-09-05 13:31 ` Ingo Molnar
2013-09-05 17:33 ` Waiman Long
2013-09-05 17:40 ` Ingo Molnar
2013-09-03 22:41 ` Sedat Dilek
2013-09-03 23:11 ` Sedat Dilek
2013-09-08 21:45 ` Linus Torvalds
2013-09-09 0:03 ` Al Viro
2013-09-09 0:25 ` Linus Torvalds
2013-09-09 0:35 ` Al Viro
2013-09-09 0:38 ` Linus Torvalds
2013-09-09 0:57 ` Al Viro
2013-09-09 2:09 ` Ramkumar Ramachandra
2013-09-09 0:30 ` Al Viro
2013-09-09 3:32 ` Linus Torvalds
2013-09-09 4:06 ` Ramkumar Ramachandra
2013-09-09 5:44 ` Al Viro
2013-08-30 17:17 ` Peter Zijlstra
2013-08-30 17:28 ` Linus Torvalds
2013-08-30 17:33 ` Linus Torvalds
2013-08-29 15:20 ` Waiman Long
2013-08-06 3:12 ` [PATCH v7 2/4] spinlock: Enable x86 architecture to do lockless refcount update Waiman Long
2013-08-06 3:12 ` [PATCH v7 3/4] dcache: replace d_lock/d_count by d_lockcnt Waiman Long
2013-08-06 3:12 ` [PATCH v7 4/4] dcache: Enable lockless update of dentry's refcount Waiman Long
2013-08-13 18:03 ` [PATCH v7 0/4] Lockless update of reference count protected by spinlock Waiman Long
2013-08-31 3:06 [PATCH v7 1/4] spinlock: A new lockref structure for lockless update of refcount George Spelvin
2013-08-31 17:16 ` Linus Torvalds
2013-09-01 8:50 ` George Spelvin
2013-09-01 11:10 ` Theodore Ts'o
2013-09-01 15:49 ` Linus Torvalds
2013-09-01 18:11 ` Steven Rostedt
2013-09-01 20:03 ` Linus Torvalds
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=5220FD51.2010709@hp.com \
--to=waiman.long@hp.com \
--cc=andi@firstfloor.org \
--cc=aswin@hp.com \
--cc=benh@kernel.crashing.org \
--cc=jlayton@redhat.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=mingo@redhat.com \
--cc=mszeredi@suse.cz \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=scott.norton@hp.com \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
--cc=viro@zeniv.linux.org.uk \
/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).