linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
From: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
To: benh@kernel.crashing.org
Cc: Fenghua Yu <fenghua.yu@intel.com>,
	Tony Luck <tony.luck@intel.com>,
	linux-ia64@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-mm@kvack.org, David Rientjes <rientjes@google.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	linuxppc-dev@lists.ozlabs.org,
	Jiang Liu <jiang.liu@linux.intel.com>,
	Wanpeng Li <liwanp@linux.vnet.ibm.com>
Subject: [RFC 0/2] Memoryless nodes and kworker
Date: Thu, 17 Jul 2014 16:09:23 -0700	[thread overview]
Message-ID: <20140717230923.GA32660@linux.vnet.ibm.com> (raw)

[Apologies for the large Cc list, but I believe we have the following
interested parties:

x86 (recently posted memoryless node support)
ia64 (existing memoryless node support)
ppc (existing memoryless node support)
previous discussion of how to solve Anton's issue with slab usage
workqueue contributors/maintainers]

There is an issue currently where NUMA information is used on powerpc
(and possibly ia64) before it has been read from the device-tree, which
leads to large slab consumption with CONFIG_SLUB and memoryless nodes.

While testing memoryless nodes on PowerKVM guests with the patches in
this series, with a guest topology of
    
    available: 2 nodes (0-1)
    node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
    node 0 size: 0 MB
    node 0 free: 0 MB
    node 1 cpus: 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
    node 1 size: 16336 MB
    node 1 free: 15329 MB
    node distances:
    node   0   1
      0:  10  40
      1:  40  10
    
the slab consumption decreases from
    
    Slab:             932416 kB
    SUnreclaim:       902336 kB
    
to
    
    Slab:             395264 kB
    SUnreclaim:       359424 kB
    
And we see a corresponding increase in the slab efficiency from
    
    slab                                   mem     objs    slabs
                                          used   active   active
    ------------------------------------------------------------
    kmalloc-16384                       337 MB   11.28%  100.00%
    task_struct                         288 MB    9.93%  100.00%
    
to
    
    slab                                   mem     objs    slabs
                                          used   active   active
    ------------------------------------------------------------
    kmalloc-16384                        37 MB  100.00%  100.00%
    task_struct                          31 MB  100.00%  100.00%

It turns out we see this large slab usage due to using the wrong NUMA
information when creating kthreads.
    
Two changes are required, one of which is in the workqueue code and one
of which is in the powerpc initialization. Note that ia64 may want to
consider something similar.

             reply	other threads:[~2014-07-17 23:09 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-07-17 23:09 Nishanth Aravamudan [this message]
2014-07-17 23:09 ` [RFC 1/2] workqueue: use the nearest NUMA node, not the local one Nishanth Aravamudan
2014-07-17 23:15   ` [RFC 2/2] powerpc: reorder per-cpu NUMA information's initialization Nishanth Aravamudan
2014-07-18  8:11   ` [RFC 1/2] workqueue: use the nearest NUMA node, not the local one Lai Jiangshan
2014-07-18 17:33     ` Nish Aravamudan
2014-07-18 11:20 ` [RFC 0/2] Memoryless nodes and kworker Tejun Heo
2014-07-18 17:42   ` Nish Aravamudan
2014-07-18 18:00     ` Tejun Heo
2014-07-18 18:01       ` Tejun Heo
2014-07-18 18:12       ` Nish Aravamudan
2014-07-18 18:19         ` Tejun Heo
2014-07-18 18:47           ` Nish Aravamudan
2014-07-18 18:58             ` Tejun Heo

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=20140717230923.GA32660@linux.vnet.ibm.com \
    --to=nacc@linux.vnet.ibm.com \
    --cc=benh@kernel.crashing.org \
    --cc=fenghua.yu@intel.com \
    --cc=iamjoonsoo.kim@lge.com \
    --cc=jiang.liu@linux.intel.com \
    --cc=linux-ia64@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=linuxppc-dev@lists.ozlabs.org \
    --cc=liwanp@linux.vnet.ibm.com \
    --cc=rientjes@google.com \
    --cc=tony.luck@intel.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;
as well as URLs for NNTP newsgroup(s).