Re: [PATCH] Introduce lifetime write IO statistics

[Chao Yu <chao2.yu@samsung.com>]

> -----Original Message-----
> From: Jaegeuk Kim [mailto:jaegeuk@kernel.org]
> Sent: Thursday, January 28, 2016 6:27 AM
> To: Chao Yu
> Cc: 'Shuoran'; linux-f2fs-devel@lists.sourceforge.net
> Subject: Re: [f2fs-dev] [PATCH] Introduce lifetime write IO statistics
> 
> Hi Chao,
> 
> On Wed, Jan 27, 2016 at 06:02:37PM +0800, Chao Yu wrote:
> > > -----Original Message-----
> > > From: Shuoran [mailto:liushuoran@huawei.com]
> > > Sent: Wednesday, January 27, 2016 3:40 PM
> > > To: Chao Yu
> > > Cc: jaegeuk@kernel.org; linux-f2fs-devel@lists.sourceforge.net
> > > Subject: Re: [f2fs-dev] [PATCH] Introduce lifetime write IO statistics
> > >
> > > On 2016/1/27 15:20, Chao Yu wrote:
> > > >> -----Original Message-----
> > > >> From: Shuoran [mailto:liushuoran@huawei.com]
> > > >> Sent: Tuesday, January 26, 2016 6:12 PM
> > > >> To: Chao Yu
> > > >> Cc: jaegeuk@kernel.org; linux-f2fs-devel@lists.sourceforge.net
> > > >> Subject: Re: [f2fs-dev] [PATCH] Introduce lifetime write IO statistics
> > > >>
> > > >> On 2016/1/26 17:24, Chao Yu wrote:
> > > >>> Hi,
> > > >>>
> > > >>>> -----Original Message-----
> > > >>>> From: Liu Shuoran [mailto:liushuoran@huawei.com]
> > > >>>> Sent: Tuesday, January 26, 2016 3:41 PM
> > > >>>> To: jaegeuk@kernel.org
> > > >>>> Cc: linux-f2fs-devel@lists.sourceforge.net
> > > >>>> Subject: [f2fs-dev] [PATCH] Introduce lifetime write IO statistics
> > > >>>>
> > > >>>> This patch introduces lifetime IO write statistics exposed to the sysfs interface.
> > > >>>> The write IO amount is obtained from block layer, accumulated in the file system and
> > > >>>> stored in the hot node summary of checkpoint.
> > > >>>>
> > > >>>> Signed-off-by: Shuoran Liu <liushuoran@huawei.com>
> > > >>>> Signed-off-by: Pengyang Hou <houpengyang@huawei.com>
> > > >>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > > >>>> ---
> > > >>>>    fs/f2fs/checkpoint.c    | 12 ++++++++++++
> > > >>>>    fs/f2fs/f2fs.h          | 11 +++++++++++
> > > >>>>    fs/f2fs/super.c         | 35 +++++++++++++++++++++++++++++++++++
> > > >>>>    include/linux/f2fs_fs.h | 14 +++++++++++++-
> > > >>>>    4 files changed, 71 insertions(+), 1 deletion(-)
> > > >>>>
> > > >>>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> > > >>>> index 3842af9..998de47 100644
> > > >>>> --- a/fs/f2fs/checkpoint.c
> > > >>>> +++ b/fs/f2fs/checkpoint.c
> > > >>>> @@ -921,6 +921,10 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct
> cp_control
> > > >> *cpc)
> > > >>>>    	int cp_payload_blks = __cp_payload(sbi);
> > > >>>>    	block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
> > > >>>>    	bool invalidate = false;
> > > >>>> +	struct super_block *sb = sbi->sb;
> > > >>>> +	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
> > > >>>> +	u64 kbytes_written;
> > > >>>> +
> > > >>>>
> > > >>>>    	/*
> > > >>>>    	 * This avoids to conduct wrong roll-forward operations and uses
> > > >>>> @@ -1034,6 +1038,14 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct
> cp_control
> > > >>>> *cpc)
> > > >>>>
> > > >>>>    	write_data_summaries(sbi, start_blk);
> > > >>>>    	start_blk += data_sum_blocks;
> > > >>>> +
> > > >>>> +	/* Record write statistics in the hot node summary */
> > > >>>> +	kbytes_written = sbi->kbytes_written;
> > > >>>> +	if (sb->s_bdev->bd_part)
> > > >>>> +		kbytes_written += BD_PART_WRITTEN(sbi);
> > > >>>> +
> > > >>>> +	seg_i->sum_blk->info.kbytes_written = cpu_to_le64(kbytes_written);
> > > >>>> +
> > > >>>>    	if (__remain_node_summaries(cpc->reason)) {
> > > >>>>    		write_node_summaries(sbi, start_blk);
> > > >>>>    		start_blk += NR_CURSEG_NODE_TYPE;
> > > >>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > >>>> index ff79054..cee5fab0 100644
> > > >>>> --- a/fs/f2fs/f2fs.h
> > > >>>> +++ b/fs/f2fs/f2fs.h
> > > >>>> @@ -844,8 +844,19 @@ struct f2fs_sb_info {
> > > >>>>    	struct list_head s_list;
> > > >>>>    	struct mutex umount_mutex;
> > > >>>>    	unsigned int shrinker_run_no;
> > > >>>> +
> > > >>>> +	/* For write statistics */
> > > >>>> +	u64 sectors_written_start;
> > > >>>> +	u64 kbytes_written;
> > > >>>>    };
> > > >>>>
> > > >>>> +/* For write statistics. Suppose sector size is 512 bytes,
> > > >>>> + * and the return value is in kbytes. s is of struct f2fs_sb_info.
> > > >>>> + */
> > > >>>> +#define BD_PART_WRITTEN(s)						 \
> > > >>>> +(((u64)part_stat_read(s->sb->s_bdev->bd_part, sectors[1]) -		 \
> > > >>>> +		s->sectors_written_start) >> 1)
> > > >>>> +
> > > >>>>    static inline void f2fs_update_time(struct f2fs_sb_info *sbi, int type)
> > > >>>>    {
> > > >>>>    	sbi->last_time[type] = jiffies;
> > > >>>> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > > >>>> index 3bf990b..7ed10f1 100644
> > > >>>> --- a/fs/f2fs/super.c
> > > >>>> +++ b/fs/f2fs/super.c
> > > >>>> @@ -126,6 +126,19 @@ static unsigned char *__struct_ptr(struct f2fs_sb_info *sbi, int
> > > >>>> struct_type)
> > > >>>>    	return NULL;
> > > >>>>    }
> > > >>>>
> > > >>>> +static ssize_t lifetime_write_kbytes_show(struct f2fs_attr *a,
> > > >>>> +		struct f2fs_sb_info *sbi, char *buf)
> > > >>>> +{
> > > >>>> +	struct super_block *sb = sbi->sb;
> > > >>>> +
> > > >>>> +	if (!sb->s_bdev->bd_part)
> > > >>>> +		return snprintf(buf, PAGE_SIZE, "0\n");
> > > >>>> +
> > > >>>> +	return snprintf(buf, PAGE_SIZE, "%llu\n",
> > > >>>> +		(unsigned long long)(sbi->kbytes_written +
> > > >>>> +			BD_PART_WRITTEN(sbi)));
> > > >>>> +}
> > > >>>> +
> > > >>>>    static ssize_t f2fs_sbi_show(struct f2fs_attr *a,
> > > >>>>    			struct f2fs_sb_info *sbi, char *buf)
> > > >>>>    {
> > > >>>> @@ -204,6 +217,9 @@ static struct f2fs_attr f2fs_attr_##_name = {			\
> > > >>>>    		f2fs_sbi_show, f2fs_sbi_store,			\
> > > >>>>    		offsetof(struct struct_name, elname))
> > > >>>>
> > > >>>> +#define F2FS_GENERAL_RO_ATTR(name) \
> > > >>>> +static struct f2fs_attr f2fs_attr_##name = __ATTR(name, 0444, name##_show, NULL)
> > > >>>> +
> > > >>>>    F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_min_sleep_time, min_sleep_time);
> > > >>>>    F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_max_sleep_time, max_sleep_time);
> > > >>>>    F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_no_gc_sleep_time, no_gc_sleep_time);
> > > >>>> @@ -220,6 +236,7 @@ F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, max_victim_search,
> > > >>>> max_victim_search);
> > > >>>>    F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, dir_level, dir_level);
> > > >>>>    F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, cp_interval, interval_time[CP_TIME]);
> > > >>>>    F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, idle_interval, interval_time[REQ_TIME]);
> > > >>>> +F2FS_GENERAL_RO_ATTR(lifetime_write_kbytes);
> > > >>>>
> > > >>>>    #define ATTR_LIST(name) (&f2fs_attr_##name.attr)
> > > >>>>    static struct attribute *f2fs_attrs[] = {
> > > >>>> @@ -239,6 +256,7 @@ static struct attribute *f2fs_attrs[] = {
> > > >>>>    	ATTR_LIST(ra_nid_pages),
> > > >>>>    	ATTR_LIST(cp_interval),
> > > >>>>    	ATTR_LIST(idle_interval),
> > > >>>> +	ATTR_LIST(lifetime_write_kbytes),
> > > >>>>    	NULL,
> > > >>>>    };
> > > >>>>
> > > >>>> @@ -766,6 +784,11 @@ static int f2fs_remount(struct super_block *sb, int *flags, char
> *data)
> > > >>>>    	bool need_stop_gc = false;
> > > >>>>    	bool no_extent_cache = !test_opt(sbi, EXTENT_CACHE);
> > > >>>>
> > > >>>> +	if (*flags & MS_RDONLY) {
> > > >>>> +		set_opt(sbi, FASTBOOT);
> > > >>> Need to recover to original mode, otherwise it may change option user
> > > >>> set.
> > > >> Yes, maybe I can bring forward the action of saving old mount options.
> > > >>>> +		set_sbi_flag(sbi, SBI_IS_DIRTY);
> > > >>>> +	}
> > > >>>> +
> > > >>>>    	sync_filesystem(sb);
> > > >>>>
> > > >>>>    	/*
> > > >>>> @@ -1242,6 +1265,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data,
> int
> > > >> silent)
> > > >>>>    	bool retry = true, need_fsck = false;
> > > >>>>    	char *options = NULL;
> > > >>>>    	int recovery, i, valid_super_block;
> > > >>>> +	struct curseg_info *seg_i;
> > > >>>>
> > > >>>>    try_onemore:
> > > >>>>    	err = -EINVAL;
> > > >>>> @@ -1372,6 +1396,17 @@ try_onemore:
> > > >>>>    		goto free_nm;
> > > >>>>    	}
> > > >>>>
> > > >>>> +	/* For write statistics */
> > > >>>> +	if (sb->s_bdev->bd_part)
> > > >>>> +		sbi->sectors_written_start =
> > > >>>> +			(u64)part_stat_read(sb->s_bdev->bd_part, sectors[1]);
> > > >>>> +
> > > >>>> +	/* Read accumulated write IO statistics if exists */
> > > >>>> +	seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
> > > >>> Why not CURSEG_WARM_DATA?
> > > >> Data summary might be compacted. So only hot/warm/cold node summary is
> > > >> available, and we just picked hot node summary.
> > > > Oh, you're right, warm data journal would not be persisted if data
> > > > summaries are compacted, my mistake.
> > > >
> > > > Wouldn't be better to persist this stat number into super block like
> > > > ext4?
> > > >
> > > > Thanks,
> > >
> > > I think super block in F2FS is not updated as frequent as that in Ext4
> > > (there are several places invoking ext4_commit_super, while few invoking
> > > f2fs_commit_super), although in this design the statistics is not
> > > updated very frequently either. Another concern is that, to some extent,
> > > the super blocks of f2fs is not supposed to update except for a few
> > > cases. So the stat is better stored in the checkpoint, and we just tried
> > > very hard finding a spare place.
> >
> > Got it, so the key point is preventing generating additional IO.
> >
> > If update points of stat in your approach is enough, so for
> > 1) umount case: commit sb approach causes two more blocks updating.
> > 2) remount case: commit sb approach causes one less blocks updating when
> > FASTBOOT is disabled.
> >
> > Not a big degression. :)
> >
> > IMO, maybe it's better to use journal log in data segments to store
> > checkpoint related data, maybe more nats/sits entries.
> >
> > And another thing is I'm in doubt that whether it is enough to use cp +
> > data journal to update stat, since abnormal pow-cut after a normal cp would
> > make us losing all stats saved before.
> 
> Pros and Cons.
> IMO, we don't need to guarantee up-to-date information to users. If users wants
> to keep that in high accuracy, they simply can set fastboot.

Yes, at least it needs to document this sysfs entry to let user know about this.

Thanks,

> 
> Thanks,
> 
> >
> > Thanks,
> >
> > >
> > > >>> Thanks,
> > > >>>
> > > >>>> +	if(__exist_node_summaries(sbi))
> > > >>>> +		sbi->kbytes_written =
> > > >>>> +			le64_to_cpu(seg_i->sum_blk->info.kbytes_written);
> > > >>>> +
> > > >>>>    	build_gc_manager(sbi);
> > > >>>>
> > > >>>>    	/* get an inode for node space */
> > > >>>> diff --git a/include/linux/f2fs_fs.h b/include/linux/f2fs_fs.h
> > > >>>> index e59c3be..67aa01d 100644
> > > >>>> --- a/include/linux/f2fs_fs.h
> > > >>>> +++ b/include/linux/f2fs_fs.h
> > > >>>> @@ -358,6 +358,12 @@ struct summary_footer {
> > > >>>>    				sizeof(struct sit_journal_entry))
> > > >>>>    #define SIT_JOURNAL_RESERVED	((SUM_JOURNAL_SIZE - 2) %\
> > > >>>>    				sizeof(struct sit_journal_entry))
> > > >>>> +
> > > >>>> +/* Reserved area should make size of f2fs_extra_info equals to
> > > >>>> + * that of nat_journal and sit_journal.
> > > >>>> + */
> > > >>>> +#define EXTRA_INFO_RESERVED	(SUM_JOURNAL_SIZE - 2 - 8)
> > > >>>> +
> > > >>>>    /*
> > > >>>>     * frequently updated NAT/SIT entries can be stored in the spare area in
> > > >>>>     * summary blocks
> > > >>>> @@ -387,6 +393,11 @@ struct sit_journal {
> > > >>>>    	__u8 reserved[SIT_JOURNAL_RESERVED];
> > > >>>>    } __packed;
> > > >>>>
> > > >>>> +struct f2fs_extra_info {
> > > >>>> +	__le64 kbytes_written;
> > > >>>> +	__u8 reserved[EXTRA_INFO_RESERVED];
> > > >>>> +} __packed;
> > > >>>> +
> > > >>>>    /* 4KB-sized summary block structure */
> > > >>>>    struct f2fs_summary_block {
> > > >>>>    	struct f2fs_summary entries[ENTRIES_IN_SUM];
> > > >>>> @@ -394,10 +405,11 @@ struct f2fs_summary_block {
> > > >>>>    		__le16 n_nats;
> > > >>>>    		__le16 n_sits;
> > > >>>>    	};
> > > >>>> -	/* spare area is used by NAT or SIT journals */
> > > >>>> +	/* spare area is used by NAT or SIT journals or extra info */
> > > >>>>    	union {
> > > >>>>    		struct nat_journal nat_j;
> > > >>>>    		struct sit_journal sit_j;
> > > >>>> +		struct f2fs_extra_info info;
> > > >>>>    	};
> > > >>>>    	struct summary_footer footer;
> > > >>>>    } __packed;
> > > >>>> --
> > > >>>> 1.9.1
> > > >>>>
> > > >>>>
> > > >>>> ------------------------------------------------------------------------------
> > > >>>> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> > > >>>> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> > > >>>> Monitor end-to-end web transactions and take corrective actions now
> > > >>>> Troubleshoot faster and improve end-user experience. Signup Now!
> > > >>>> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
> > > >>>> _______________________________________________
> > > >>>> Linux-f2fs-devel mailing list
> > > >>>> Linux-f2fs-devel@lists.sourceforge.net
> > > >>>> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> > > >>> .
> > > >>>
> > > >
> > > >
> > > > .
> > > >
> > >
> >


------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140