Re: [PATCH] f2fs: avoid issuing small bios due to several dirty node pages

[Jaegeuk Kim <jaegeuk.kim@samsung.com>]

[-- Attachment #1: Type: text/plain, Size: 2723 bytes --]

2013-01-18 (금), 17:57 +0900, Namjae Jeon:
> 2013/1/18, Jaegeuk Kim <jaegeuk.kim@samsung.com>:
> > 2013-01-18 (금), 16:58 +0900, Namjae Jeon:
> >> 2013/1/18, Jaegeuk Kim <jaegeuk.kim@samsung.com>:
> >> > If some small bios of dirty node pages are supposed to be issued during
> >> > the
> >> > sequential data writes, there-in well-produced consecutive data bios
> >> > are
> >> > able
> >> > to be split by the small node bios, resulting in performance
> >> > degradation.
> >> > So, let's collect a number of dirty node pages until reaching a
> >> > threshold.
> >> > And, by default, I set the threshold as 2MB, a segment size.
> >> >
> >> > This improves sequential write performance on i5, 512GB SSD (830 w/
> >> > SATA2)
> >> > as
> >> > follows.
> >> > Before: 231 MB/s -> After: 255 MB/s
> >> >
> >> > Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
> >> > ---
> >> >  fs/f2fs/node.c | 17 +++++++++++------
> >> >  1 file changed, 11 insertions(+), 6 deletions(-)
> >> >
> >> > diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> >> > index f177c01..9bda63c 100644
> >> > --- a/fs/f2fs/node.c
> >> > +++ b/fs/f2fs/node.c
> >> > @@ -1124,6 +1124,12 @@ static int f2fs_write_node_page(struct page
> >> > *page,
> >> >  	return 0;
> >> >  }
> >> >
> >> > +/*
> >> > + * It is very important to gather dirty pages and write at once, so
> >> > that we
> >> > can
> >> > + * submit a big bio without interfering other data writes.
> >> > + * Be default, 512 pages (2MB), a segment size, is quite reasonable.
> >> > + */
> >> > +#define COLLECT_DIRTY_NODES	512
> >> Hi Jaeguek.
> >> It is just my opinion.
> >> One of f2fs advantages is that user can set segment size to fit own
> >> device.
> >> For future, Is it not good to use segment size set by user(from
> >> superblock) instead of fixed size ?
> >
> > At this moment, it's very hard to change the segment size due to many
> > hardcoded data structures such as SSA, bitmaps, etc.
> > Instead, I proposed something like section and zone which are based on
> > segment.
> > But, somebody will be able to do later. :)
> Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
> Okay, I think It is better if you add TODO comment about this to be
> fixed by the other.
> 
> Talk Incidentally, let me ask you other thing.
> get_meta_page never return NULL about allocate page and readpage.
> Is there any reason it try to infinitely retry about two case ?

That is just to handle errors more easily especially for file system
metadata operations.
Definitely, we need to consider all the error handling codes in more
detail by using the fault-injection facility later.
Thanks,

-- 
Jaegeuk Kim
Samsung

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]