Re: regression in page writeback

[Wu Fengguang <fengguang.wu@intel.com>]

On Wed, Sep 23, 2009 at 11:14:50AM +0800, Wu Fengguang wrote:
> On Wed, Sep 23, 2009 at 11:10:12AM +0800, Li, Shaohua wrote:
> > On Wed, Sep 23, 2009 at 10:49:58AM +0800, Wu, Fengguang wrote:
> > > On Wed, Sep 23, 2009 at 10:36:22AM +0800, Andrew Morton wrote:
> > > > On Wed, 23 Sep 2009 10:26:22 +0800 Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > > 
> > > > > On Wed, Sep 23, 2009 at 09:59:41AM +0800, Andrew Morton wrote:
> > > > > > On Wed, 23 Sep 2009 09:45:00 +0800 Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > > > > 
> > > > > > > On Wed, Sep 23, 2009 at 09:28:32AM +0800, Andrew Morton wrote:
> > > > > > > > On Wed, 23 Sep 2009 09:17:58 +0800 Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > > > > > > 
> > > > > > > > > On Wed, Sep 23, 2009 at 08:54:52AM +0800, Andrew Morton wrote:
> > > > > > > > > > On Wed, 23 Sep 2009 08:22:20 +0800 Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > > > > > > > > 
> > > > > > > > > > > Jens' per-bdi writeback has another improvement. In 2.6.31, when
> > > > > > > > > > > superblocks A and B both have 100000 dirty pages, it will first
> > > > > > > > > > > exhaust A's 100000 dirty pages before going on to sync B's.
> > > > > > > > > > 
> > > > > > > > > > That would only be true if someone broke 2.6.31.  Did they?
> > > > > > > > > > 
> > > > > > > > > > SYSCALL_DEFINE0(sync)
> > > > > > > > > > {
> > > > > > > > > > 	wakeup_pdflush(0);
> > > > > > > > > > 	sync_filesystems(0);
> > > > > > > > > > 	sync_filesystems(1);
> > > > > > > > > > 	if (unlikely(laptop_mode))
> > > > > > > > > > 		laptop_sync_completion();
> > > > > > > > > > 	return 0;
> > > > > > > > > > }
> > > > > > > > > > 
> > > > > > > > > > the sync_filesystems(0) is supposed to non-blockingly start IO against
> > > > > > > > > > all devices.  It used to do that correctly.  But people mucked with it
> > > > > > > > > > so perhaps it no longer does.
> > > > > > > > > 
> > > > > > > > > I'm referring to writeback_inodes(). Each invocation of which (to sync
> > > > > > > > > 4MB) will do the same iteration over superblocks A => B => C ... So if
> > > > > > > > > A has dirty pages, it will always be served first.
> > > > > > > > > 
> > > > > > > > > So if wbc->bdi == NULL (which is true for kupdate/background sync), it
> > > > > > > > > will have to first exhaust A before going on to B and C.
> > > > > > > > 
> > > > > > > > But that works OK.  We fill the first device's queue, then it gets
> > > > > > > > congested and sync_sb_inodes() does nothing and we advance to the next
> > > > > > > > queue.
> > > > > > > 
> > > > > > > So in common cases "exhaust" is a bit exaggerated, but A does receive
> > > > > > > much more opportunity than B. Computation resources for IO submission
> > > > > > > are unbalanced for A, and there are pointless overheads in rechecking A.
> > > > > > 
> > > > > > That's unquantified handwaving.  One CPU can do a *lot* of IO.
> > > > > 
> > > > > Yes.. I had the impression that the writeback submission can be pretty slow.
> > > > > It should be because of the congestion_wait. Now that it is removed,
> > > > > things are going faster when queue is not full.
> > > > 
> > > > What?  The wait is short.  The design intent there is that we repoll
> > > > all previously-congested queues well before they start to run empty.
> > > 
> > > When queue is not congested (in which case congestion_wait is not
> > > necessary), the congestion_wait() degrades io submission speed to near
> > > io completion speed.
> > > 
> > > > > > > > If a device has more than a queue's worth of dirty data then we'll
> > > > > > > > probably leave some of that dirty memory un-queued, so there's some
> > > > > > > > lack of concurrency in that situation.
> > > > > > > 
> > > > > > > Good insight.
> > > > > > 
> > > > > > It was wrong.  See the other email.
> > > > > 
> > > > > No your first insight is correct. Because the (unnecessary) teeny
> > > > > sleeps is independent of the A=>B=>C traversing order. Only queue
> > > > > congestion could help skip A.
> > > > 
> > > > The sleeps are completely necessary!  Otherwise we end up busywaiting.
> > > > 
> > > > After the sleep we repoll all queues.
> > > 
> > > I mean, it is not always necessary. Only when _all_ superblocks cannot
> > > writeback their inodes (eg. all in congestion), we should wait.
> > > 
> > > Just before Jens' work, I had patch to convert
> > > 
> > > -                       if (wbc.encountered_congestion || wbc.more_io)
> > > -                               congestion_wait(WRITE, HZ/10);
> > > -                       else
> > > -                               break;
> > > 
> > > to
> > > 
> > > +       if (wbc->encountered_congestion && wbc->nr_to_write == MAX_WRITEBACK_PAGES)
> > > +               congestion_wait(WRITE, HZ/10);
> > > 
> > > Note that wbc->encountered_congestion only means "at least one bdi
> > > encountered congestion". We may still make progress in other bdis
> > > hence should not sleep.
> > Hi,
> > encountered_congestion only is checked when nr_to_write > 0, if some superblocks
> > aren't congestions, nr_to_write should be 0, right?
> 
> Yeah, good spot! So the change only helps some corner cases.

Then it remains a problem why the io submission is slow when !congested.

For example, this trace shows that the io submission speed is about

        4MB / 0.01s = 400MB/s

Workload is a plain copy on a 2Ghz Intel Core 2 CPU.

[   71.487121] mm/page-writeback.c 761 background_writeout: comm=pdflush pid=343 n=-4096
[   71.489635] global dirty=65513 writeback=5925 nfs=1 flags=_M towrite=0 skipped=0
[   71.496019] redirty_tail +442: inode 79232
[   71.497432] mm/page-writeback.c 761 background_writeout: comm=pdflush pid=343 n=-5120
[   71.498890] global dirty=64490 writeback=6700 nfs=1 flags=_M towrite=0 skipped=0
[   71.506355] redirty_tail +442: inode 79232
[   71.508473] mm/page-writeback.c 761 background_writeout: comm=pdflush pid=343 n=-6144
[   71.510538] global dirty=62475 writeback=7599 nfs=1 flags=_M towrite=0 skipped=0
[   71.511910] redirty_tail +502: inode 3438
[   71.512846] redirty_tail +502: inode 1920

Thanks,
Fengguang