[PATCH v1] f2fs: New victim selection for GC

From: Yonggil Song
Date: Thu Oct 12 2023 - 06:04:10 EST

Next message: Abel Wu: "Re: Re: [PATCH] sched/fair: fix pick_eevdf to always find the correct se"
Previous message: Dan Carpenter: "Re: [PATCH] staging/wlan-ng: remove strcpy() use in favor of strscpy()"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Overview
========

Introduce a new way to select the data section first when selecting a
victim in foreground GC. This victim selection method works when the
prefer_data_victim mount option is enabled. If foreground GC migrates only
data sections and runs out of free sections, it cleans dirty node sections
to get more free sections.

Problem
=======

If the total amount of nodes is larger than the size of one section, nodes
occupy multiple sections, and node victims are often selected because the
gc cost is lowered by data block migration in foreground gc. Since moving
the data section causes frequent node victim selection, victim threshing
occurs in the node section. This results in an increase in WAF.

Experiment
==========

Test environment is as follows.

System info
- 3.6GHz, 16 core CPU
- 36GiB Memory
Device info
- a conventional null_blk with 228MiB
- a sequential null_blk with 4068 zones of 8MiB
Format
- mkfs.f2fs <conv null_blk> -c <seq null_blk> -m -Z 8 -o 3.89
Mount
- mount -o prefer_data_victim <conv null_blk> <mount point>
Fio script
- fio --rw=randwrite --bs=4k --ba=4k --filesize=31187m --norandommap --overwrite=1 --name=job1 --filename=./mnt/sustain --io_size=128g
WAF calculation
- (IOs on conv. null_blk + IOs on seq. null_blk) / random write IOs

Conclusion
==========

This experiment showed that the WAF was reduced by 29% (18.75 -> 13.3) when
the data section was selected first when selecting GC victims. This was
achieved by reducing the migration of the node blocks by 69.4%
(253,131,743 blks -> 77,463,278 blks). It is possible to achieve low WAF
performance with the GC victim selection method in environments where the
section size is relatively small.

Signed-off-by: Yonggil Song <yonggil.song@xxxxxxxxxxx>
---
Documentation/filesystems/f2fs.rst | 3 +
fs/f2fs/f2fs.h | 2 +
fs/f2fs/gc.c | 100 +++++++++++++++++++++++------
fs/f2fs/segment.h | 2 +
fs/f2fs/super.c | 9 +++
5 files changed, 95 insertions(+), 21 deletions(-)

diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index d32c6209685d..58e6d001d7ab 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -367,6 +367,9 @@ errors=%s Specify f2fs behavior on critical errors. This supports modes:
pending node write drop keep N/A
pending meta write keep keep N/A
====================== =============== =============== ========
+prefer_data_victim When selecting victims in foreground GC, victims of data type
+ are prioritized. This option minimizes GC victim threshing
+ in the node section to reduce WAF.
======================== ============================================================

Debugfs Entries
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 6d688e42d89c..8b31fa2ea09a 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -108,6 +108,7 @@ extern const char *f2fs_fault_name[FAULT_MAX];
#define F2FS_MOUNT_GC_MERGE 0x02000000
#define F2FS_MOUNT_COMPRESS_CACHE 0x04000000
#define F2FS_MOUNT_AGE_EXTENT_CACHE 0x08000000
+#define F2FS_MOUNT_PREFER_DATA_VICTIM 0x10000000

#define F2FS_OPTION(sbi) ((sbi)->mount_opt)
#define clear_opt(sbi, option) (F2FS_OPTION(sbi).opt &= ~F2FS_MOUNT_##option)
@@ -1648,6 +1649,7 @@ struct f2fs_sb_info {
struct f2fs_mount_info mount_opt; /* mount options */

/* for cleaning operations */
+ bool need_node_clean; /* only used for prefer_data_victim */
struct f2fs_rwsem gc_lock; /*
* semaphore for GC, avoid
* race between GC and GC or CP
diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index f550cdeaa663..8a2da808a5fb 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -752,6 +752,8 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
unsigned int last_segment;
unsigned int nsearched;
bool is_atgc;
+ bool is_prefer_data_victim =
+ test_opt(sbi, PREFER_DATA_VICTIM) && gc_type == FG_GC;
int ret = 0;

mutex_lock(&dirty_i->seglist_lock);
@@ -767,6 +769,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
p.oldest_age = 0;
p.min_cost = get_max_cost(sbi, &p);

+ if (is_prefer_data_victim) {
+ p.node_min_cost = p.min_cost;
+ p.node_min_segno = p.min_segno;
+ }
+
is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
nsearched = 0;

@@ -884,9 +891,25 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,

cost = get_gc_cost(sbi, segno, &p);

- if (p.min_cost > cost) {
- p.min_segno = segno;
- p.min_cost = cost;
+ if (is_prefer_data_victim) {
+ if (IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
+ /* update data segments victim */
+ if (p.min_cost > cost) {
+ p.min_segno = segno;
+ p.min_cost = cost;
+ }
+ } else {
+ /* update node segments victim */
+ if (p.node_min_cost > cost) {
+ p.node_min_segno = segno;
+ p.node_min_cost = cost;
+ }
+ }
+ } else {
+ if (p.min_cost > cost) {
+ p.min_segno = segno;
+ p.min_cost = cost;
+ }
}
next:
if (nsearched >= p.max_search) {
@@ -901,6 +924,25 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
}
}

+ if (is_prefer_data_victim && sbi->need_node_clean) {
+ /* we need to clean node sections */
+ if (p.min_cost > p.node_min_cost) {
+ p.min_segno = p.node_min_segno;
+ p.min_cost = p.node_min_cost;
+ } else {
+ /*
+ * data victim cost is the lowest.
+ * if free sections are enough, stop cleaning node victim.
+ * if not, it goes on by GCing data victims.
+ */
+ if (has_enough_free_secs(sbi, prefree_segments(sbi), 0)) {
+ sbi->need_node_clean = false;
+ p.min_segno = NULL_SEGNO;
+ goto out;
+ }
+ }
+ }
+
/* get victim for GC_AT/AT_SSR */
if (is_atgc) {
lookup_victim_by_age(sbi, &p);
@@ -1830,8 +1872,27 @@ int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
goto stop;
}

+ __get_secs_required(sbi, NULL, &upper_secs, NULL);
+
+ /*
+ * Write checkpoint to reclaim prefree segments.
+ * We need more three extra sections for writer's data/node/dentry.
+ */
+ if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS) {
+ if (test_opt(sbi, PREFER_DATA_VICTIM)) {
+ sbi->need_node_clean = true;
+ }
+ if (prefree_segments(sbi)) {
+ ret = f2fs_write_checkpoint(sbi, &cpc);
+ if (ret)
+ goto stop;
+ /* Reset due to checkpoint */
+ sec_freed = 0;
+ }
+ }
+
/* Let's run FG_GC, if we don't have enough space. */
- if (has_not_enough_free_secs(sbi, 0, 0)) {
+ if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
gc_type = FG_GC;

/*
@@ -1882,7 +1943,17 @@ int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
if (!gc_control->no_bg_gc &&
total_sec_freed < gc_control->nr_free_secs)
goto go_gc_more;
- goto stop;
+ if (test_opt(sbi, PREFER_DATA_VICTIM)) {
+ /*
+ * If the need_node_clean flag is set
+ * even though there are enough free
+ * sections, node cleaning will continue.
+ */
+ if (!sbi->need_node_clean)
+ goto stop;
+ } else {
+ goto stop;
+ }
}
if (sbi->skipped_gc_rwsem)
skipped_round++;
@@ -1897,21 +1968,6 @@ int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
goto stop;
}

- __get_secs_required(sbi, NULL, &upper_secs, NULL);
-
- /*
- * Write checkpoint to reclaim prefree segments.
- * We need more three extra sections for writer's data/node/dentry.
- */
- if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS &&
- prefree_segments(sbi)) {
- stat_inc_cp_call_count(sbi, TOTAL_CALL);
- ret = f2fs_write_checkpoint(sbi, &cpc);
- if (ret)
- goto stop;
- /* Reset due to checkpoint */
- sec_freed = 0;
- }
go_gc_more:
segno = NULL_SEGNO;
goto gc_more;
@@ -1920,8 +1976,10 @@ int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;

- if (gc_type == FG_GC)
+ if (gc_type == FG_GC) {
f2fs_unpin_all_sections(sbi, true);
+ sbi->need_node_clean = false;
+ }

trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed,
get_pages(sbi, F2FS_DIRTY_NODES),
diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
index 2ca8fb5d0dc4..d55fa1fee2e0 100644
--- a/fs/f2fs/segment.h
+++ b/fs/f2fs/segment.h
@@ -197,8 +197,10 @@ struct victim_sel_policy {
unsigned int offset; /* last scanned bitmap offset */
unsigned int ofs_unit; /* bitmap search unit */
unsigned int min_cost; /* minimum cost */
+ unsigned int node_min_cost; /* minimum cost of node type section */
unsigned long long oldest_age; /* oldest age of segments having the same min cost */
unsigned int min_segno; /* segment # having min. cost */
+ unsigned int node_min_segno; /* node segment # having min. cost */
unsigned long long age; /* mtime of GCed section*/
unsigned long long age_threshold;/* age threshold */
};
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index a8c8232852bb..133137dd6fd0 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -165,6 +165,7 @@ enum {
Opt_memory_mode,
Opt_age_extent_cache,
Opt_errors,
+ Opt_prefer_data_victim,
Opt_err,
};

@@ -245,6 +246,7 @@ static match_table_t f2fs_tokens = {
{Opt_memory_mode, "memory=%s"},
{Opt_age_extent_cache, "age_extent_cache"},
{Opt_errors, "errors=%s"},
+ {Opt_prefer_data_victim, "prefer_data_victim"},
{Opt_err, NULL},
};

@@ -1286,6 +1288,13 @@ static int parse_options(struct super_block *sb, char *options, bool is_remount)
}
kfree(name);
break;
+ case Opt_prefer_data_victim:
+ if (!f2fs_sb_has_blkzoned(sbi)) {
+ f2fs_err(sbi, "prefer_data_victim is only allowed with zoned block device feature");
+ return -EINVAL;
+ }
+ set_opt(sbi, PREFER_DATA_VICTIM);
+ break;
default:
f2fs_err(sbi, "Unrecognized mount option \"%s\" or missing value",
p);
--
2.34.1

Next message: Abel Wu: "Re: Re: [PATCH] sched/fair: fix pick_eevdf to always find the correct se"
Previous message: Dan Carpenter: "Re: [PATCH] staging/wlan-ng: remove strcpy() use in favor of strscpy()"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]