From: Yang Guo <[email protected]>
@es_stats_cache_hits and @es_stats_cache_misses are accessed frequently in
ext4_es_lookup_extent function, it would influence the ext4 read/write
performance in NUMA system. Let's optimize it using percpu_counter,
it is profitable for the performance.
The test command is as below:
fio -name=randwrite -numjobs=8 -filename=/mnt/test1 -rw=randwrite
-ioengine=libaio -direct=1 -iodepth=64 -sync=0 -norandommap
-group_reporting -runtime=120 -time_based -bs=4k -size=5G
And the result is better 10% than the initial implement:
without the patch，IOPS=197k, BW=770MiB/s (808MB/s)(90.3GiB/120002msec)
with the patch, IOPS=218k, BW=852MiB/s (894MB/s)(99.9GiB/120002msec)
Cc: "Theodore Ts'o" <[email protected]>
Cc: Andreas Dilger <[email protected]>
Cc: Eric Biggers <[email protected]>
Signed-off-by: Yang Guo <[email protected]>
Signed-off-by: Shaokun Zhang <[email protected]>
---
ChangeLog:
Fix the issue that there is no percpu_counter_destroy() for the new
percpu counters.
fs/ext4/extents_status.c | 37 ++++++++++++++++++++++++-------------
fs/ext4/extents_status.h | 4 ++--
2 files changed, 26 insertions(+), 15 deletions(-)
diff --git a/fs/ext4/extents_status.c b/fs/ext4/extents_status.c
index 7521de2dcf3a..3c03062a8d6a 100644
--- a/fs/ext4/extents_status.c
+++ b/fs/ext4/extents_status.c
@@ -947,9 +947,9 @@ int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
es->es_pblk = es1->es_pblk;
if (!ext4_es_is_referenced(es1))
ext4_es_set_referenced(es1);
- stats->es_stats_cache_hits++;
+ percpu_counter_inc(&stats->es_stats_cache_hits);
} else {
- stats->es_stats_cache_misses++;
+ percpu_counter_inc(&stats->es_stats_cache_misses);
}
read_unlock(&EXT4_I(inode)->i_es_lock);
@@ -1235,9 +1235,9 @@ int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
- seq_printf(seq, " %lu/%lu cache hits/misses\n",
- es_stats->es_stats_cache_hits,
- es_stats->es_stats_cache_misses);
+ seq_printf(seq, " %lld/%lld cache hits/misses\n",
+ percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
+ percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
if (inode_cnt)
seq_printf(seq, " %d inodes on list\n", inode_cnt);
@@ -1264,35 +1264,46 @@ int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
sbi->s_es_nr_inode = 0;
spin_lock_init(&sbi->s_es_lock);
sbi->s_es_stats.es_stats_shrunk = 0;
- sbi->s_es_stats.es_stats_cache_hits = 0;
- sbi->s_es_stats.es_stats_cache_misses = 0;
+ err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
+ GFP_KERNEL);
+ if (err)
+ return err;
+ err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
+ GFP_KERNEL);
+ if (err)
+ goto err1;
sbi->s_es_stats.es_stats_scan_time = 0;
sbi->s_es_stats.es_stats_max_scan_time = 0;
err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
if (err)
- return err;
+ goto err2;
err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
if (err)
- goto err1;
+ goto err3;
sbi->s_es_shrinker.scan_objects = ext4_es_scan;
sbi->s_es_shrinker.count_objects = ext4_es_count;
sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
err = register_shrinker(&sbi->s_es_shrinker);
if (err)
- goto err2;
+ goto err4;
return 0;
-
-err2:
+err4:
percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
-err1:
+err3:
percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
+err2:
+ percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
+err1:
+ percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
return err;
}
void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
{
+ percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
+ percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
unregister_shrinker(&sbi->s_es_shrinker);
diff --git a/fs/ext4/extents_status.h b/fs/ext4/extents_status.h
index 131a8b7df265..e722dd9bd06e 100644
--- a/fs/ext4/extents_status.h
+++ b/fs/ext4/extents_status.h
@@ -70,8 +70,8 @@ struct ext4_es_tree {
struct ext4_es_stats {
unsigned long es_stats_shrunk;
- unsigned long es_stats_cache_hits;
- unsigned long es_stats_cache_misses;
+ struct percpu_counter es_stats_cache_hits;
+ struct percpu_counter es_stats_cache_misses;
u64 es_stats_scan_time;
u64 es_stats_max_scan_time;
struct percpu_counter es_stats_all_cnt;
--
2.7.4
On Wed, Aug 28, 2019 at 05:19:17PM +0800, Shaokun Zhang wrote:
> From: Yang Guo <[email protected]>
>
> @es_stats_cache_hits and @es_stats_cache_misses are accessed frequently in
> ext4_es_lookup_extent function, it would influence the ext4 read/write
> performance in NUMA system. Let's optimize it using percpu_counter,
> it is profitable for the performance.
>
> The test command is as below:
> fio -name=randwrite -numjobs=8 -filename=/mnt/test1 -rw=randwrite
> -ioengine=libaio -direct=1 -iodepth=64 -sync=0 -norandommap
> -group_reporting -runtime=120 -time_based -bs=4k -size=5G
>
> And the result is better 10% than the initial implement:
> without the patch，IOPS=197k, BW=770MiB/s (808MB/s)(90.3GiB/120002msec)
> with the patch, IOPS=218k, BW=852MiB/s (894MB/s)(99.9GiB/120002msec)
>
> Cc: "Theodore Ts'o" <[email protected]>
> Cc: Andreas Dilger <[email protected]>
> Cc: Eric Biggers <[email protected]>
> Signed-off-by: Yang Guo <[email protected]>
> Signed-off-by: Shaokun Zhang <[email protected]>
Thanks, applied.
- Ted