RE: [f2fs-dev] [PATCH] f2fs: reduce competition among node page writes

[Chao Yu]

Hi Jaegeuk,

> -----Original Message-----
> From: Jaegeuk Kim [mailto:jaegeuk@xxxxxxxxxx]
> Sent: Thursday, July 31, 2014 2:45 PM
> To: Chao Yu
> Cc: 'Changman Lee'; linux-fsdevel@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> Subject: Re: [f2fs-dev] [PATCH] f2fs: reduce competition among node page writes
> 
> On Thu, Jul 31, 2014 at 01:31:46PM +0800, Chao Yu wrote:
> > Hi Changman,
> >
> > > -----Original Message-----
> > > From: Changman Lee [mailto:cm224.lee@xxxxxxxxxxx]
> > > Sent: Thursday, July 31, 2014 10:07 AM
> > > To: Chao Yu
> > > Cc: 'Jaegeuk Kim'; linux-fsdevel@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> > > linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > > Subject: Re: [f2fs-dev] [PATCH] f2fs: reduce competition among node page writes
> > >
> > > Hi Chao,
> > >
> > > On Wed, Jul 30, 2014 at 09:07:49PM +0800, Chao Yu wrote:
> > > > Hi Jaegeuk Changman,
> > > >
> > > > > -----Original Message-----
> > > > > From: Chao Yu [mailto:chao2.yu@xxxxxxxxxxx]
> > > > > Sent: Thursday, July 03, 2014 6:59 PM
> > > > > To: Jaegeuk Kim; Changman Lee
> > > > > Cc: linux-fsdevel@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> > > > > linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > > > > Subject: [f2fs-dev] [PATCH] f2fs: reduce competition among node page writes
> > > > >
> > > > > We do not need to block on ->node_write among different node page writers e.g.
> > > > > fsync/flush, unless we have a node page writer from write_checkpoint.
> > > > > So it's better use rw_semaphore instead of mutex type for ->node_write to
> > > > > promote performance.
> > > >
> > > > If you could have time to help explaining the problem of this patch, I will be
> > > > appreciated for that.
> > >
> > > I have no clue. Except checkpoint, I don't know why need to block to
> > > write node page.
> > > Do you have any problem when you test with this patch?
> >
> > I don't have.
> > I send this patch about one month ago, but got no respond.
> > So I want to ask if any problem in this patch or forget to look at this patch?
> >
> > To Jaegeuk:
> > Any idea about this patch?
> 
> Oh, I forgot to send an email for this.
> At that time, when I looked at a glance, I thought that it was not clear why
> this should be merged.
> 
> But, when I contemplate again, it seems that several fsync threads could produce
> multiple node writers, resulting in some mutex contention.
> Just for sure, can you verify that?

Yes, node sync in cp could encounter competition of the same op in fsync/flush/gc
thread. Here we use rwlock to increase concurrent of these thread hence we could
gain better performance of checkpoint.

Thanks,
Yu

> 
> Nevertheless, I think there would be no problem to merge this patch now.
> Merged.
> 
> >
> > >
> > > >
> > > > Another question is what is ->writepages in sbi used for? I'm not quite clear.
> > > >
> > >
> > > I remember it is for writing data pages per thread as much as possible.
> > > When multi-threads write some files simultaneously, multi-threads contended with
> > > each other to allocate a block. So block allocation was interleaved
> > > across threads. It makes fragmentation of file.
> 
> Good. :)
> 
> >
> > Thank you for the explanation! :)
> > I think what you say is reasonable.
> >
> > Previously I tested without this lock, although I found that the blocks written
> > _almost_ were continuous in each '->writepages()'. Still I think we can gain more
> > from readahead continuous block when using this lock, rather than remove it for
> > promoting concurrent of writers.
> >
> > Thanks,
> > Yu
> >
> > >
> > > Thanks,
> > >
> > > > Thanks,
> > > >
> > > > >
> > > > > Signed-off-by: Chao Yu <chao2.yu@xxxxxxxxxxx>
> > > > > ---
> > > > >  fs/f2fs/checkpoint.c |    6 +++---
> > > > >  fs/f2fs/f2fs.h       |    2 +-
> > > > >  fs/f2fs/node.c       |    4 ++--
> > > > >  fs/f2fs/super.c      |    2 +-
> > > > >  4 files changed, 7 insertions(+), 7 deletions(-)
> > > > >
> > > > > diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> > > > > index 0b4710c..eec406b 100644
> > > > > --- a/fs/f2fs/checkpoint.c
> > > > > +++ b/fs/f2fs/checkpoint.c
> > > > > @@ -714,10 +714,10 @@ retry_flush_dents:
> > > > >  	 * until finishing nat/sit flush.
> > > > >  	 */
> > > > >  retry_flush_nodes:
> > > > > -	mutex_lock(&sbi->node_write);
> > > > > +	down_write(&sbi->node_write);
> > > > >
> > > > >  	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
> > > > > -		mutex_unlock(&sbi->node_write);
> > > > > +		up_write(&sbi->node_write);
> > > > >  		sync_node_pages(sbi, 0, &wbc);
> > > > >  		goto retry_flush_nodes;
> > > > >  	}
> > > > > @@ -726,7 +726,7 @@ retry_flush_nodes:
> > > > >
> > > > >  static void unblock_operations(struct f2fs_sb_info *sbi)
> > > > >  {
> > > > > -	mutex_unlock(&sbi->node_write);
> > > > > +	up_write(&sbi->node_write);
> > > > >  	f2fs_unlock_all(sbi);
> > > > >  }
> > > > >
> > > > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > > > index ae3b4ac..ca30b5a 100644
> > > > > --- a/fs/f2fs/f2fs.h
> > > > > +++ b/fs/f2fs/f2fs.h
> > > > > @@ -444,7 +444,7 @@ struct f2fs_sb_info {
> > > > >  	struct inode *meta_inode;		/* cache meta blocks */
> > > > >  	struct mutex cp_mutex;			/* checkpoint procedure lock */
> > > > >  	struct rw_semaphore cp_rwsem;		/* blocking FS operations */
> > > > > -	struct mutex node_write;		/* locking node writes */
> > > > > +	struct rw_semaphore node_write;		/* locking node writes */
> > > > >  	struct mutex writepages;		/* mutex for writepages() */
> > > > >  	bool por_doing;				/* recovery is doing or not */
> > > > >  	wait_queue_head_t cp_wait;
> > > > > diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> > > > > index a90f51d..7b5b5de 100644
> > > > > --- a/fs/f2fs/node.c
> > > > > +++ b/fs/f2fs/node.c
> > > > > @@ -1231,12 +1231,12 @@ static int f2fs_write_node_page(struct page *page,
> > > > >  	if (wbc->for_reclaim)
> > > > >  		goto redirty_out;
> > > > >
> > > > > -	mutex_lock(&sbi->node_write);
> > > > > +	down_read(&sbi->node_write);
> > > > >  	set_page_writeback(page);
> > > > >  	write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
> > > > >  	set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
> > > > >  	dec_page_count(sbi, F2FS_DIRTY_NODES);
> > > > > -	mutex_unlock(&sbi->node_write);
> > > > > +	up_read(&sbi->node_write);
> > > > >  	unlock_page(page);
> > > > >  	return 0;
> > > > >
> > > > > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > > > > index 8f96d93..bed9413 100644
> > > > > --- a/fs/f2fs/super.c
> > > > > +++ b/fs/f2fs/super.c
> > > > > @@ -947,7 +947,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int
> silent)
> > > > >  	mutex_init(&sbi->gc_mutex);
> > > > >  	mutex_init(&sbi->writepages);
> > > > >  	mutex_init(&sbi->cp_mutex);
> > > > > -	mutex_init(&sbi->node_write);
> > > > > +	init_rwsem(&sbi->node_write);
> > > > >  	sbi->por_doing = false;
> > > > >  	spin_lock_init(&sbi->stat_lock);
> > > > >
> > > > > --
> > > > > 1.7.9.5
> > > > >
> > > > >
> > > > >
> > > > > ------------------------------------------------------------------------------
> > > > > Open source business process management suite built on Java and Eclipse
> > > > > Turn processes into business applications with Bonita BPM Community Edition
> > > > > Quickly connect people, data, and systems into organized workflows
> > > > > Winner of BOSSIE, CODIE, OW2 and Gartner awards
> > > > > http://p.sf.net/sfu/Bonitasoft
> > > > > _______________________________________________
> > > > > Linux-f2fs-devel mailing list
> > > > > Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > > > > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/