LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Andrew Morton <akpm@osdl.org>
To: Christoph Lameter <clameter@sgi.com>
Cc: menage@google.com, linux-kernel@vger.kernel.org,
	nickpiggin@yahoo.com.au, linux-mm@kvack.org, ak@suse.de,
	pj@sgi.com, dgc@sgi.com
Subject: Re: [RFC 0/8] Cpuset aware writeback
Date: Tue, 16 Jan 2007 15:40:54 -0800	[thread overview]
Message-ID: <20070116154054.e655f75c.akpm@osdl.org> (raw)
In-Reply-To: <Pine.LNX.4.64.0701161407530.3545@schroedinger.engr.sgi.com>

> On Tue, 16 Jan 2007 14:15:56 -0800 (PST) Christoph Lameter <clameter@sgi.com> wrote:
>
> ...
>
> > > This may result in a large percentage of a cpuset
> > > to become dirty without writeout being triggered. Under NFS
> > > this can lead to OOM conditions.
> > 
> > OK, a big question: is this patchset a performance improvement or a
> > correctness fix?  Given the above, and the lack of benchmark results I'm
> > assuming it's for correctness.
> 
> It is a correctness fix both for NFS OOM and doing proper cpuset writeout.

It's a workaround for a still-unfixed NFS problem.

> > - Why does NFS go oom?  Because it allocates potentially-unbounded
> >   numbers of requests in the writeback path?
> > 
> >   It was able to go oom on non-numa machines before dirty-page-tracking
> >   went in.  So a general problem has now become specific to some NUMA
> >   setups.
> 
> 
> Right. The issue is that large portions of memory become dirty / 
> writeback since no writeback occurs because dirty limits are not checked 
> for a cpuset. Then NFS attempt to writeout when doing LRU scans but is 
> unable to allocate memory.
>  
> >   So an obvious, equivalent and vastly simpler "fix" would be to teach
> >   the NFS client to go off-cpuset when trying to allocate these requests.
> 
> Yes we can fix these allocations by allowing processes to allocate from 
> other nodes. But then the container function of cpusets is no longer 
> there.

But that's what your patch already does!

It asks pdflush to write the pages instead of the direct-reclaim caller. 
The only reason pdflush doesn't go oom is that pdflush lives outside the
direct-reclaim caller's cpuset and is hence able to obtain those nfs
requests from off-cpuset zones.

> > (But is it really bad? What actual problems will it cause once NFS is fixed?)
> 
> NFS is okay as far as I can tell. dirty throttling works fine in non 
> cpuset environments because we throttle if 40% of memory becomes dirty or 
> under writeback.

Repeat: NFS shouldn't go oom.  It should fail the allocation, recover, wait
for existing IO to complete.  Back that up with a mempool for NFS requests
and the problem is solved, I think?

> > I don't understand why the proposed patches are cpuset-aware at all.  This
> > is a per-zone problem, and a per-zone fix would seem to be appropriate, and
> > more general.  For example, i386 machines can presumably get into trouble
> > if all of ZONE_DMA or ZONE_NORMAL get dirty.  A good implementation would
> > address that problem as well.  So I think it should all be per-zone?
> 
> No. A zone can be completely dirty as long as we are allowed to allocate 
> from other zones.

But we also can get into trouble if a *zone* is all-dirty.  Any solution to
the cpuset problem should solve that problem too, no?

> > Do we really need those per-inode cpumasks?  When page reclaim encounters a
> > dirty page on the zone LRU, we automatically know that page->mapping->host
> > has at least one dirty page in this zone, yes?  We could immediately ask
> 
> Yes, but when we enter reclaim most of the pages of a zone may already be 
> dirty/writeback so we fail.

No.  If the dirty limits become per-zone then no zone will ever have >40%
dirty.

The obvious fix here is: when a zone hits 40% dirty, perform dirty-memory
reduction in that zone, throttling the dirtying process.  I suspect this
would work very badly in common situations with, say, typical i386 boxes.



  reply	other threads:[~2007-01-16 23:41 UTC|newest]

Thread overview: 110+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-01-16  5:47 Christoph Lameter
2007-01-16  5:47 ` [RFC 1/8] Convert higest_possible_node_id() into nr_node_ids Christoph Lameter
2007-01-16 22:05   ` Andi Kleen
2007-01-17  3:14     ` Christoph Lameter
2007-01-17  4:15       ` Andi Kleen
2007-01-17  4:23         ` Christoph Lameter
2007-01-16  5:47 ` [RFC 2/8] Add a map to inodes to track dirty pages per node Christoph Lameter
2007-01-16  5:47 ` [RFC 3/8] Add a nodemask to pdflush functions Christoph Lameter
2007-01-16  5:48 ` [RFC 4/8] Per cpuset dirty ratio handling and writeout Christoph Lameter
2007-01-16  5:48 ` [RFC 5/8] Make writeout during reclaim cpuset aware Christoph Lameter
2007-01-16 22:07   ` Andi Kleen
2007-01-17  4:20     ` Paul Jackson
2007-01-17  4:28       ` Andi Kleen
2007-01-17  4:36         ` Paul Jackson
2007-01-17  5:59           ` Andi Kleen
2007-01-17  6:19             ` Christoph Lameter
2007-01-17  4:23     ` Christoph Lameter
2007-01-16  5:48 ` [RFC 6/8] Throttle vm writeout per cpuset Christoph Lameter
2007-01-16  5:48 ` [RFC 7/8] Exclude unreclaimable pages from dirty ration calculation Christoph Lameter
2007-01-18 15:48   ` Nikita Danilov
2007-01-18 19:56     ` Christoph Lameter
2007-01-16  5:48 ` [RFC 8/8] Reduce inode memory usage for systems with a high MAX_NUMNODES Christoph Lameter
2007-01-16 19:52   ` Paul Menage
2007-01-16 20:00     ` Christoph Lameter
2007-01-16 20:06       ` Paul Menage
2007-01-16 20:51         ` Christoph Lameter
2007-01-16  7:38 ` [RFC 0/8] Cpuset aware writeback Peter Zijlstra
2007-01-16 20:10   ` Christoph Lameter
2007-01-16  9:25 ` Paul Jackson
2007-01-16 17:13   ` Christoph Lameter
2007-01-16 21:53 ` Andrew Morton
2007-01-16 22:08   ` [PATCH] nfs: fix congestion control Peter Zijlstra
2007-01-16 22:27     ` Trond Myklebust
2007-01-17  2:41       ` Peter Zijlstra
2007-01-17  6:15         ` Trond Myklebust
2007-01-17  8:49           ` Peter Zijlstra
2007-01-17 13:50             ` Trond Myklebust
2007-01-17 14:29               ` Peter Zijlstra
2007-01-17 14:45                 ` Trond Myklebust
2007-01-17 20:05     ` Christoph Lameter
2007-01-17 21:52       ` Peter Zijlstra
2007-01-17 21:54         ` Trond Myklebust
2007-01-18 13:27           ` Peter Zijlstra
2007-01-18 15:49             ` Trond Myklebust
2007-01-19  9:33               ` Peter Zijlstra
2007-01-19 13:07                 ` Peter Zijlstra
2007-01-19 16:51                   ` Trond Myklebust
2007-01-19 17:54                     ` Peter Zijlstra
2007-01-19 17:20                   ` Christoph Lameter
2007-01-19 17:57                     ` Peter Zijlstra
2007-01-19 18:02                       ` Christoph Lameter
2007-01-19 18:26                       ` Trond Myklebust
2007-01-19 18:27                         ` Christoph Lameter
2007-01-20  7:01                         ` [PATCH] nfs: fix congestion control -v3 Peter Zijlstra
2007-01-22 16:12                           ` Trond Myklebust
2007-01-25 15:32                             ` [PATCH] nfs: fix congestion control -v4 Peter Zijlstra
2007-01-26  5:02                               ` Andrew Morton
2007-01-26  8:00                                 ` Peter Zijlstra
2007-01-26  8:50                                   ` Peter Zijlstra
2007-01-26  5:09                               ` Andrew Morton
2007-01-26  5:31                                 ` Christoph Lameter
2007-01-26  6:04                                   ` Andrew Morton
2007-01-26  6:53                                     ` Christoph Lameter
2007-01-26  8:03                                     ` Peter Zijlstra
2007-01-26  8:51                                       ` Andrew Morton
2007-01-26  9:01                                         ` Peter Zijlstra
2007-02-20 12:59                                         ` Peter Zijlstra
2007-01-22 17:59                           ` [PATCH] nfs: fix congestion control -v3 Christoph Lameter
2007-01-17 23:15     ` [PATCH] nfs: fix congestion control Christoph Hellwig
2007-01-16 22:15   ` [RFC 0/8] Cpuset aware writeback Christoph Lameter
2007-01-16 23:40     ` Andrew Morton [this message]
2007-01-17  0:16       ` Christoph Lameter
2007-01-17  1:07         ` Andrew Morton
2007-01-17  1:30           ` Christoph Lameter
2007-01-17  2:34             ` Andrew Morton
2007-01-17  3:40               ` Christoph Lameter
2007-01-17  4:02                 ` Paul Jackson
2007-01-17  4:05                 ` Andrew Morton
2007-01-17  6:27                   ` Christoph Lameter
2007-01-17  7:00                     ` Andrew Morton
2007-01-17  8:01                       ` Paul Jackson
2007-01-17  9:57                         ` Andrew Morton
2007-01-17 19:43                       ` Christoph Lameter
2007-01-17 22:10                         ` Andrew Morton
2007-01-18  1:10                           ` Christoph Lameter
2007-01-18  1:25                             ` Andrew Morton
2007-01-18  5:21                               ` Christoph Lameter
2007-01-16 23:44   ` David Chinner
2007-01-16 22:01 ` Andi Kleen
2007-01-16 22:18   ` Christoph Lameter
2007-02-02  1:38 ` Ethan Solomita
2007-02-02  2:16   ` Christoph Lameter
2007-02-02  4:03     ` Andrew Morton
2007-02-02  5:29       ` Christoph Lameter
2007-02-02  6:02         ` Neil Brown
2007-02-02  6:17           ` Christoph Lameter
2007-02-02  6:41             ` Neil Brown
2007-02-02  7:12         ` Andrew Morton
2007-03-21 21:11     ` Ethan Solomita
2007-03-21 21:29       ` Christoph Lameter
2007-03-21 21:52         ` Andrew Morton
2007-03-21 21:57           ` Christoph Lameter
2007-04-19  2:07         ` Ethan Solomita
2007-04-19  2:55           ` Christoph Lameter
2007-04-19  7:52             ` Ethan Solomita
2007-04-19 16:03               ` Christoph Lameter
2007-04-21  1:37             ` Ethan Solomita
2007-04-21  1:48               ` Christoph Lameter
2007-04-21  8:15                 ` Ethan Solomita
2007-04-21 15:40                   ` Christoph Lameter

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=20070116154054.e655f75c.akpm@osdl.org \
    --to=akpm@osdl.org \
    --cc=ak@suse.de \
    --cc=clameter@sgi.com \
    --cc=dgc@sgi.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=menage@google.com \
    --cc=nickpiggin@yahoo.com.au \
    --cc=pj@sgi.com \
    --subject='Re: [RFC 0/8] Cpuset aware writeback' \
    /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

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).