Hi,
On Tue, Dec 23, 2014 at 06:04:54PM +0900, Changman Lee wrote:
> Hi,
>
> On Tue, Dec 23, 2014 at 12:00:37AM -0800, Jaegeuk Kim wrote:
> > Hi Changman,
> >
> > On Tue, Dec 23, 2014 at 08:37:38AM +0900, Changman Lee wrote:
> > > f2fs has 2 gc_type; foreground gc and background gc.
> > > In the case of foreground gc, f2fs will select victim as greedy.
> > > Otherwise, as cost-benefit. And also it runs as greedy in SSR mode.
> > > Until now, f2fs_gc conducted with BG_GC as default. So we couldn't
> > > expect how it runs; BG_GC or FG_GC and GREEDY or COST_BENEFIT.
> >
> > What does this mean?
> > In f2fs_gc, the gc_type will be changed accoring to the number of free
> > sections.
>
> Right, but when I turn on trace I saw 3 cases.
> 1. BG_GC and COST_BENEFIT
> 2. BG_GC and GREEDY
> 3. FG_GC and GREEDY
>
> I expected that case 1 is likely to operate only by gc_thread.
> But it was not.
One of FG_GC and BG_GC is selected according to the # of free segments.
One of CB and GREEDY is determined by select_policy().
There are orthogonal.
Please, do not expect #1 only by gc_thread.
We don't need to force it by callers, since the policy should be selected
by f2fs_gc to avoid frequent FG_GC unnecessarily.
>
> >
> > > Therefore sometimes it runs as BG_GC/COST_BENEFIT although gc_thread
> > > don't put f2fs_gc to work.
> >
> > You mean f2fs_balance_fs?
> > In this case, again, the gc_type will be assigned FG_GC.
> >
> > Why do you want to set FG_GC/GREEDY for the SSR victims?
>
> We should alloate a block as soon as possible.
> In the case of FG_GC, it also uses invalid blocks dirtied by background gc.
Yes, because FG_GC needs to move victim blocks.
> In another case, if (BG_GC && test_bit(victim_secmap)), it will be
> skipped.
> I intended that SSR operates fastly like FG_GC.
If FG_GC is used by SSR, we lose the effect on pre-loaded victim blocks done by
BG_GC before.
This must be a good trial to enhance SSR latencies.
Just in case, can you show how much this is serious?
If so, we need to think about different ways.
Thanks,
>
> Regards,
> Changman
>
> >
> > Thanks,
> >
> > >
> > > Signed-off-by: Changman Lee <cm224.lee@xxxxxxxxxxx>
> > > ---
> > > fs/f2fs/f2fs.h | 2 +-
> > > fs/f2fs/gc.c | 5 ++---
> > > fs/f2fs/segment.c | 6 +++---
> > > 3 files changed, 6 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > index ae6dfb6..c956535 100644
> > > --- a/fs/f2fs/f2fs.h
> > > +++ b/fs/f2fs/f2fs.h
> > > @@ -1476,7 +1476,7 @@ int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *, u64, u64);
> > > int start_gc_thread(struct f2fs_sb_info *);
> > > void stop_gc_thread(struct f2fs_sb_info *);
> > > block_t start_bidx_of_node(unsigned int, struct f2fs_inode_info *);
> > > -int f2fs_gc(struct f2fs_sb_info *);
> > > +int f2fs_gc(struct f2fs_sb_info *, int);
> > > void build_gc_manager(struct f2fs_sb_info *);
> > > int __init create_gc_caches(void);
> > > void destroy_gc_caches(void);
> > > diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> > > index eec0933..e1fa53a 100644
> > > --- a/fs/f2fs/gc.c
> > > +++ b/fs/f2fs/gc.c
> > > @@ -80,7 +80,7 @@ static int gc_thread_func(void *data)
> > > stat_inc_bggc_count(sbi);
> > >
> > > /* if return value is not zero, no victim was selected */
> > > - if (f2fs_gc(sbi))
> > > + if (f2fs_gc(sbi, BG_GC))
> > > wait_ms = gc_th->no_gc_sleep_time;
> > >
> > > /* balancing f2fs's metadata periodically */
> > > @@ -691,10 +691,9 @@ static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
> > > f2fs_put_page(sum_page, 1);
> > > }
> > >
> > > -int f2fs_gc(struct f2fs_sb_info *sbi)
> > > +int f2fs_gc(struct f2fs_sb_info *sbi, int gc_type)
> > > {
> > > unsigned int segno, i;
> > > - int gc_type = BG_GC;
> > > int nfree = 0;
> > > int ret = -1;
> > > struct cp_control cpc;
> > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > index fd9bc96..3b32404 100644
> > > --- a/fs/f2fs/segment.c
> > > +++ b/fs/f2fs/segment.c
> > > @@ -281,7 +281,7 @@ void f2fs_balance_fs(struct f2fs_sb_info *sbi)
> > > */
> > > if (has_not_enough_free_secs(sbi, 0)) {
> > > mutex_lock(&sbi->gc_mutex);
> > > - f2fs_gc(sbi);
> > > + f2fs_gc(sbi, FG_GC);
> > > }
> > > }
> > >
> > > @@ -994,12 +994,12 @@ static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
> > >
> > > if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
> > > return v_ops->get_victim(sbi,
> > > - &(curseg)->next_segno, BG_GC, type, SSR);
> > > + &(curseg)->next_segno, FG_GC, type, SSR);
> > >
> > > /* For data segments, let's do SSR more intensively */
> > > for (; type >= CURSEG_HOT_DATA; type--)
> > > if (v_ops->get_victim(sbi, &(curseg)->next_segno,
> > > - BG_GC, type, SSR))
> > > + FG_GC, type, SSR))
> > > return 1;
> > > return 0;
> > > }
> > > --
> > > 1.9.1
> > >
> > >
> > > ------------------------------------------------------------------------------
> > > Dive into the World of Parallel Programming! The Go Parallel Website,
> > > sponsored by Intel and developed in partnership with Slashdot Media, is your
> > > hub for all things parallel software development, from weekly thought
> > > leadership blogs to news, videos, case studies, tutorials and more. Take a
> > > look and join the conversation now. http://goparallel.sourceforge.net
> > > _______________________________________________
> > > 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-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
