mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: "Martin J. Bligh" <mbligh@aracnet.com>
To: Andrea Arcangeli <andrea@suse.de>, linux-kernel@vger.kernel.org
Cc: Hugh Dickins <hugh@veritas.com>,
	William Lee Irwin III <wli@holomorphy.com>,
	Dave McCracken <dmccr@us.ibm.com>
Subject: Re: 230-objrmap fixes for 2.6.3-mjb2
Date: Wed, 03 Mar 2004 07:54:39 -0800	[thread overview]
Message-ID: <8030000.1078329278@[10.10.2.4]> (raw)
In-Reply-To: <20040303070933.GB4922@dualathlon.random>



--Andrea Arcangeli <andrea@suse.de> wrote (on Wednesday, March 03, 2004 08:09:33 +0100):

> While merging 230-objrmap in my tree I spotted 2 bugs potentially
> generating random memory corruption and 1 superflous bit that I dropped
> (mostly for documentation reasons, I like strict and in turn self
> documenting). Here below the fixes.

Looks good to me, but Dave is more familiar with this stuff ... Dave?

> I'm running some shm swap regression test on this right now and I'll
> leave it running for a day. In a few hours I will proceed starting
> dropping the pte_chain from the page sturcture and then I'll test the
> anon swapout. I will also follow the 6 great-effort anobjrmap posted by
> Hugh against objrmap while doing that, they're quite old (almost 1 year)
> but they still apply cleanly by hand so they're useful.

Bill has been keeping that up to date  - he may have some more recent
changes somewhere?

> About 2.5:1.5 it seems not everybody is happy to lose 512m (and it's not
> Oracle), but before ruling it out I'd like to get some real life number,
> to be sure the performance of 2.0^W4:4 are really close (if not
> "better") than 3:1 as someone said. If we go with 4:4 IMHO at the very
> least the vgettimeofday backport from x86-64 is a must. In the meantime

John has that going - there's a copy in 2.6.3-mjb1, but it has a problem
at the moment on some boxes, which he's trying to fix up. I've been 
pushing for the same thing - fixing the most common syscall will help ;-)
We're trying to get some numbers here, but have had a small setback with
some disk problems - should be this week still though.

> I keep going with the rmap removal to fixup the fork  and to get back
> the 128m of normal zone useful on the 32G boxes. Could be also that new
> cpus are a lot better at reloading the tlbs from the pagetables dunno,
> the first numbers I recall about 4:4 predates to 2000 when PII was quite
> optimal.  I'd only like to see an opteron and a xeon dealing with 4:4.

Makes sense. But why would you want 4:4 on opteron? I'm not sure we care
about the perf for anyone nutty enough to run a 32 bit kernel there ;-)

M.

  parent reply	other threads:[~2004-03-03 15:59 UTC|newest]

Thread overview: 28+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2004-03-03  7:09 Andrea Arcangeli
2004-03-03 10:58 ` Andrew Morton
2004-03-03 15:46   ` Martin J. Bligh
2004-03-03 15:58     ` Dave McCracken
2004-03-03 16:08       ` Christoph Hellwig
2004-03-03 19:04         ` Mike Kravetz
2004-03-04 15:41         ` Rik van Riel
2004-03-04 15:47           ` Christoph Hellwig
2004-03-04 16:21             ` Rik van Riel
2004-03-04 17:03           ` Matt Mackall
2004-03-03 16:57     ` Andrea Arcangeli
2004-03-03 17:07       ` Dave McCracken
2004-03-03 18:39         ` Andrea Arcangeli
2004-03-03 18:44           ` Dave McCracken
2004-03-03 18:51             ` Andrea Arcangeli
2004-03-03 19:01               ` Dave McCracken
2004-03-03 21:05                 ` Andrea Arcangeli
2004-03-03 21:39               ` Martin J. Bligh
2004-03-03 23:16                 ` Andrea Arcangeli
2004-03-03 17:06   ` Andrea Arcangeli
2004-03-03 15:54 ` Martin J. Bligh [this message]
2004-03-03 16:14   ` Andrea Arcangeli
2004-03-25 21:45 ` Andrea Arcangeli
2004-03-26 11:57   ` Hugh Dickins
2004-03-26 18:02     ` Andrea Arcangeli
2004-03-26 23:25       ` Andrew Morton
2004-03-27 14:24         ` Andrea Arcangeli
2004-03-27 15:29         ` Dave McCracken

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='8030000.1078329278@[10.10.2.4]' \
    --to=mbligh@aracnet.com \
    --cc=andrea@suse.de \
    --cc=dmccr@us.ibm.com \
    --cc=hugh@veritas.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=wli@holomorphy.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

all inboxes | Powered by JetHome®