From: Paul Jackson <pj@sgi.com>
To: Ingo Molnar <mingo@elte.hu>
Cc: akpm@osdl.org, dgc@sgi.com, steiner@sgi.com, Simon.Derr@bull.net,
ak@suse.de, linux-kernel@vger.kernel.org, clameter@sgi.com
Subject: Re: [PATCH 1/5] cpuset memory spread basic implementation
Date: Sun, 5 Feb 2006 23:22:53 -0800 [thread overview]
Message-ID: <20060205232253.ddbf02d7.pj@sgi.com> (raw)
In-Reply-To: <20060206061743.GA14679@elte.hu>
> I'm wondering whether it would be
> enough to simply extend madvise and fadvise to 'task' scope as well, and
> change the pagecache allocation pattern to 'spread out' pages on NUMA,
> if POSIX_FADV_RANDOM / MADV_RANDOM is specified.
This would not seem to work, at least for the needs I am aware of.
The tasks we are talking about do -not- want a default RANDOM
policy. They want node-local allocation for per-thread data
(data and stack for example), and at the same time spread
allocation for kernel space (page and slab cache).
For another thing, memory spreading is not the same as RANDOM
policies. RANDOM policies apply to page read ahead and retention
strategies, not to page placement (what node they are faulted into)
policies.
For a third thing, madvise takes a virtual address range, which
is irrelevant for specifying kernel address space page and slab
cache pages.
But the biggest difficulty, from my perspective, would be that
this strategy is normally selected per-cpuset, not per-task.
We are managing memory placement across the cpuset, on a per-job
basis. It's the system administrator or batch scheduler, not the
individual application coder, who will likely want to enforce this
alternative memory placement strategy.
The madvise and posix_fadvise calls have no provision for one task
to affect another. They just apply to the current task. As Andrew
noted, it doesn't make much sense for different tasks in the same
cpuset to disagree on this placement policy choice. We need a cpuset
wide policy choice, and a cpuset wide mechanism for making the choice,
not a per-task task internal mechanism.
Or at the very least, an attribute that is inherited across fork
and exec, so that a job grandfather (founding father) task can
set the policy, for all descendents.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <pj@sgi.com> 1.925.600.0401
next prev parent reply other threads:[~2006-02-06 7:23 UTC|newest]
Thread overview: 102+ messages / expand[flat|nested] mbox.gz Atom feed top
2006-02-04 7:19 [PATCH 1/5] cpuset memory spread basic implementation Paul Jackson
2006-02-04 7:19 ` [PATCH 2/5] cpuset memory spread page cache implementation and hooks Paul Jackson
2006-02-04 23:49 ` Andrew Morton
2006-02-05 1:42 ` Paul Jackson
2006-02-05 1:54 ` Andrew Morton
2006-02-05 3:28 ` Christoph Lameter
2006-02-05 5:06 ` Andrew Morton
2006-02-05 6:08 ` Paul Jackson
2006-02-05 6:15 ` Andrew Morton
2006-02-05 6:28 ` Paul Jackson
2006-02-06 0:20 ` Paul Jackson
2006-02-06 5:51 ` Paul Jackson
2006-02-06 7:14 ` Pekka J Enberg
2006-02-06 7:42 ` Pekka J Enberg
2006-02-06 7:51 ` Pekka J Enberg
2006-02-06 17:32 ` Pekka Enberg
2006-02-04 7:19 ` [PATCH 3/5] cpuset memory spread slab cache implementation Paul Jackson
2006-02-04 23:49 ` Andrew Morton
2006-02-05 3:37 ` Christoph Lameter
2006-02-04 7:19 ` [PATCH 4/5] cpuset memory spread slab cache optimizations Paul Jackson
2006-02-04 23:50 ` Andrew Morton
2006-02-05 3:18 ` Paul Jackson
2006-02-04 23:50 ` Andrew Morton
2006-02-05 4:10 ` Paul Jackson
2006-02-04 7:19 ` [PATCH 5/5] cpuset memory spread slab cache hooks Paul Jackson
2006-02-06 4:37 ` Andrew Morton
2006-02-04 23:49 ` [PATCH 1/5] cpuset memory spread basic implementation Andrew Morton
2006-02-05 3:35 ` Christoph Lameter
2006-02-06 4:33 ` Andrew Morton
2006-02-06 5:50 ` Paul Jackson
2006-02-06 6:02 ` Andrew Morton
2006-02-06 6:17 ` Ingo Molnar
2006-02-06 7:22 ` Paul Jackson [this message]
2006-02-06 7:43 ` Ingo Molnar
2006-02-06 8:19 ` Paul Jackson
2006-02-06 8:22 ` Ingo Molnar
2006-02-06 8:40 ` Ingo Molnar
2006-02-06 9:03 ` Paul Jackson
2006-02-06 9:09 ` Ingo Molnar
2006-02-06 9:27 ` Paul Jackson
2006-02-06 9:37 ` Ingo Molnar
2006-02-06 20:22 ` Paul Jackson
2006-02-06 8:47 ` Paul Jackson
2006-02-06 8:51 ` Ingo Molnar
2006-02-06 9:09 ` Paul Jackson
2006-02-06 10:09 ` Andi Kleen
2006-02-06 10:11 ` Ingo Molnar
2006-02-06 10:16 ` Andi Kleen
2006-02-06 10:23 ` Ingo Molnar
2006-02-06 10:35 ` Andi Kleen
2006-02-06 14:42 ` Paul Jackson
2006-02-06 14:35 ` Paul Jackson
2006-02-06 16:48 ` Christoph Lameter
2006-02-06 17:11 ` Andi Kleen
2006-02-06 18:21 ` Christoph Lameter
2006-02-06 18:36 ` Andi Kleen
2006-02-06 18:43 ` Christoph Lameter
2006-02-06 18:48 ` Andi Kleen
2006-02-06 19:19 ` Christoph Lameter
2006-02-06 20:27 ` Paul Jackson
2006-02-06 18:43 ` Ingo Molnar
2006-02-06 20:01 ` Paul Jackson
2006-02-06 20:05 ` Ingo Molnar
2006-02-06 20:27 ` Christoph Lameter
2006-02-06 20:41 ` Ingo Molnar
2006-02-06 20:49 ` Christoph Lameter
2006-02-06 21:07 ` Ingo Molnar
2006-02-06 22:10 ` Christoph Lameter
2006-02-06 23:29 ` Ingo Molnar
2006-02-06 23:45 ` Paul Jackson
2006-02-07 0:19 ` Ingo Molnar
2006-02-07 1:17 ` David Chinner
2006-02-07 9:31 ` Andi Kleen
2006-02-07 11:53 ` Ingo Molnar
2006-02-07 12:14 ` Andi Kleen
2006-02-07 12:30 ` Ingo Molnar
2006-02-07 12:43 ` Andi Kleen
2006-02-07 12:58 ` Ingo Molnar
2006-02-07 13:14 ` Andi Kleen
2006-02-07 14:11 ` Ingo Molnar
2006-02-07 14:23 ` Andi Kleen
2006-02-07 17:11 ` Christoph Lameter
2006-02-07 17:29 ` Andi Kleen
2006-02-07 17:39 ` Christoph Lameter
2006-02-07 17:10 ` Christoph Lameter
2006-02-07 17:28 ` Andi Kleen
2006-02-07 17:42 ` Christoph Lameter
2006-02-07 17:51 ` Andi Kleen
2006-02-07 17:06 ` Christoph Lameter
2006-02-07 17:26 ` Andi Kleen
2006-02-04 23:50 ` Andrew Morton
2006-02-04 23:57 ` David S. Miller
2006-02-06 4:37 ` Andrew Morton
2006-02-06 6:02 ` Ingo Molnar
2006-02-06 6:56 ` Paul Jackson
2006-02-06 7:08 ` Andrew Morton
2006-02-06 7:39 ` Ingo Molnar
2006-02-06 8:22 ` Paul Jackson
2006-02-06 8:35 ` Paul Jackson
2006-02-06 9:32 ` Paul Jackson
2006-02-06 9:57 ` Andrew Morton
2006-02-06 9:18 ` Simon Derr
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=20060205232253.ddbf02d7.pj@sgi.com \
--to=pj@sgi.com \
--cc=Simon.Derr@bull.net \
--cc=ak@suse.de \
--cc=akpm@osdl.org \
--cc=clameter@sgi.com \
--cc=dgc@sgi.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=steiner@sgi.com \
/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