linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Christoph Lameter <clameter@sgi.com>
To: Mel Gorman <mel@skynet.ie>
Cc: akpm@linux-foundation.org, linux-mm@kvack.org
Subject: Re: [patch 09/10] SLUB: Exploit page mobility to increase allocation order
Date: Fri, 27 Apr 2007 10:15:41 -0700 (PDT)	[thread overview]
Message-ID: <Pine.LNX.4.64.0704271008390.1873@schroedinger.engr.sgi.com> (raw)
In-Reply-To: <20070427111431.GF3645@skynet.ie>

On Fri, 27 Apr 2007, Mel Gorman wrote:

> On (26/04/07 21:27), clameter@sgi.com didst pronounce:
> > If there is page mobility then we can defragment memory. So its possible to
> > use higher order of pages for slab allocations.
> > 
> > If the defaults were not overridden set the max order to 4 and guarantee 16
> > objects per slab. This will put some stress on Mel's antifrag approaches.
> > If these defaults are too large then they should be later reduced.
> > 
> 
> I see this went through mm-commits. When the next -mm kernel comes out,
> I'll grind them through the external fragmentation tests and see how it
> works out. Not all slabs are reclaimable so it might have side-effects
> if there are large amounts of slab allocations that are not allocated
> __GFP_RECLAIMABLE. Testing will tell.

Well you have not seen the whole story then. I have a draft here of a 
patch to implement slab callbacks to free objects. I think the first 
victim will be the dentry cache. I will use that functionality first to
defrag the slab cache by

1. Sort the slabs on the partial list by the number of objects inuse
   (already in mm).

2. Start from the back of the list with the smallest number of objects
   and use the callback to either free or reallocate the object. That
   will allocate new objects from the slabs with the most objects.
   Meaning the partial list will shrink on both head and tail.

3. With that I could provide you with a function to attempt to free
   up a slab page which could be used in some form for defragmentation
   from the page allocator.
   Would be great if we could work out a protocol on how to do this.
   This will initially be done with the dentry cache.

This advanced SLUB reclaim material is not suitable for 2.6.22 and I will 
keep it out of mm for awhile.

--
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>

  reply	other threads:[~2007-04-27 17:15 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-04-27  4:26 [patch 00/10] SLUB patches against 2.6.21-rc7-mm2 clameter
2007-04-27  4:26 ` [patch 01/10] SLUB: Remove duplicate VM_BUG_ON clameter
2007-04-27  4:26 ` [patch 02/10] SLUB: Fix sysfs directory handling clameter
2007-04-27  6:31   ` Andrew Morton
2007-04-27  7:02     ` Christoph Lameter
2007-04-27  7:10       ` Andrew Morton
2007-04-27  4:26 ` [patch 03/10] SLUB: debug printk cleanup clameter
2007-04-27  6:32   ` Andrew Morton
2007-04-27  4:26 ` [patch 04/10] SLUB: Conform more to SLABs SLAB_HWCACHE_ALIGN behavior clameter
2007-04-27  4:27 ` [patch 05/10] SLUB: Add MIN_PARTIAL clameter
2007-04-27  4:27 ` [patch 06/10] SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink clameter
2007-04-27  4:27 ` [patch 07/10] SLUB: Major slabinfo update clameter
2007-04-27  4:27 ` [patch 08/10] SLUB: Reduce the order of allocations to avoid fragmentation clameter
2007-04-27  4:27 ` [patch 09/10] SLUB: Exploit page mobility to increase allocation order clameter
2007-04-27  6:32   ` Andrew Morton
2007-04-27  7:04     ` Christoph Lameter
2007-04-27 11:14   ` Mel Gorman
2007-04-27 17:15     ` Christoph Lameter [this message]
2007-04-27  4:27 ` [patch 10/10] SLUB: i386 support clameter

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=Pine.LNX.4.64.0704271008390.1873@schroedinger.engr.sgi.com \
    --to=clameter@sgi.com \
    --cc=akpm@linux-foundation.org \
    --cc=linux-mm@kvack.org \
    --cc=mel@skynet.ie \
    /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