[PATCH v2] f2fs: serialize block allocation of dio writes to enhance multithread performance
From: Chao Yu
Date: Tue Dec 29 2015 - 01:44:32 EST
When performing big dio writes concurrently, our performace will be low
because of Thread A's allocation of multi continuous blocks will be
interrupted by Thread B, there are two cases as below:
- In Thread B, we may change current segment to a new segment for LFS
allocation if we dio write in the beginning of the file.
- In Thread B, we may allocate blocks in the middle of Thread A's
allocation, which make blocks allocated in Thread A being inconsecutive.
This patch adds writepages mutex lock to make block allocation in dio write
being atomic to avoid above issues.
Test environment 1:
ubuntu os with linux kernel 4.4-rc4, intel i7-3770, 16g memory,
32g kingston sd card.
fio --name seqw --ioengine=sync --invalidate=1 --rw=write --directory=/mnt/f2fs --filesize=256m --size=16m --bs=2m --direct=1
--numjobs=10
before:
WRITE: io=163840KB, aggrb=5125KB/s, minb=512KB/s, maxb=776KB/s, mint=21105msec, maxt=31967msec
patched:
WRITE: io=163840KB, aggrb=10424KB/s, minb=1042KB/s, maxb=1172KB/s, mint=13975msec, maxt=15717msec
Test environment 2:
Note4 eMMC
fio --name seqw --ioengine=sync --invalidate=1 --rw=write --directory=/data/test/ --filesize=256m --size=64m --bs=2m --direct=1
--numjobs=16
before:
WRITE: io=1024.0MB, aggrb=103583KB/s, minb=6473KB/s, maxb=8806KB/s, mint=7442msec, maxt=10123msec
patched:
WRITE: io=1024.0MB, aggrb=124860KB/s, minb=7803KB/s, maxb=9315KB/s, mint=7035msec, maxt=8398msec
As Yunlei He reported when he test with current patch:
"Does share writepages mutex lock have an effect on cache write?
Here is AndroBench result on my phone:
Before patch:
1R1W 8R8W 16R16W
Sequential Write 161.31 163.85 154.67
Random Write 9.48 17.66 18.09
After patch:
1R1W 8R8W 16R16W
Sequential Write 159.61 157.24 160.11
Random Write 9.17 8.51 8.8
Unit:Mb/s, File size: 64M, Buffer size: 4k"
The turth is androidbench uses single thread with dio write to test performance
of sequential write, and use multi-threads with dio write to test performance
of random write. so we can not see any improvement in sequentail write test
since serializing dio page allocation can only improve performance in
multi-thread scenario, and there is a regression in multi-thread test with 4k
dio write, this is because grabbing sbi->writepages lock for serializing block
allocation stop the concurrency, so that less small dio bios could be merged,
moreover, when there are huge number of small dio writes, grabbing mutex lock
per dio increases the overhead.
After all, serializing dio could only be used for concurrent scenario of
big dio, so this patch also introduces a threshold in sysfs to provide user
the interface of defining 'a big dio' with specified page number, which could
be used to control wthether serialize or not that kind of dio with specified
page number.
The optimization works in rare scenario.
Signed-off-by: Chao Yu <chao2.yu@xxxxxxxxxxx>
---
v2:
- merge another related patch into this one.
---
Documentation/ABI/testing/sysfs-fs-f2fs | 12 ++++++++++++
fs/f2fs/data.c | 17 +++++++++++++----
fs/f2fs/f2fs.h | 3 +++
fs/f2fs/super.c | 3 +++
4 files changed, 31 insertions(+), 4 deletions(-)
diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
index 0345f2d..560a4f1 100644
--- a/Documentation/ABI/testing/sysfs-fs-f2fs
+++ b/Documentation/ABI/testing/sysfs-fs-f2fs
@@ -92,3 +92,15 @@ Date: October 2015
Contact: "Chao Yu" <chao2.yu@xxxxxxxxxxx>
Description:
Controls the count of nid pages to be readaheaded.
+
+What: /sys/fs/f2fs/<disk>/serialized_dio_pages
+Date: December 2015
+Contact: "Chao Yu" <chao2.yu@xxxxxxxxxxx>
+Description:
+ It is a threshold with the unit of page size.
+ If DIO page count is equal or big than the threshold,
+ whole process of block address allocation of dio pages
+ will become atomic like buffered write.
+ It is used to maximize bandwidth utilization in the
+ scenario of concurrent write with dio vs buffered or
+ dio vs dio.
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 8a89810..d506a0e 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -1619,7 +1619,9 @@ static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
struct inode *inode = mapping->host;
+ struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
size_t count = iov_iter_count(iter);
+ int rw = iov_iter_rw(iter);
int err;
/* we don't need to use inline_data strictly */
@@ -1634,20 +1636,27 @@ static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
if (err)
return err;
- trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
+ trace_f2fs_direct_IO_enter(inode, offset, count, rw);
+
+ if (rw == WRITE) {
+ bool serialized = (F2FS_BYTES_TO_BLK(count) >=
+ sbi->serialized_dio_pages);
- if (iov_iter_rw(iter) == WRITE) {
+ if (serialized)
+ mutex_lock(&sbi->writepages);
err = __allocate_data_blocks(inode, offset, count);
+ if (serialized)
+ mutex_unlock(&sbi->writepages);
if (err)
goto out;
}
err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
out:
- if (err < 0 && iov_iter_rw(iter) == WRITE)
+ if (err < 0 && rw == WRITE)
f2fs_write_failed(mapping, offset + count);
- trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
+ trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
return err;
}
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index a339508..293dc4e 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -333,6 +333,8 @@ enum {
#define MAX_DIR_RA_PAGES 4 /* maximum ra pages of dir */
+#define DEF_SERIALIZED_DIO_PAGES 64 /* default serialized dio pages */
+
/* vector size for gang look-up from extent cache that consists of radix tree */
#define EXT_TREE_VEC_SIZE 64
@@ -784,6 +786,7 @@ struct f2fs_sb_info {
unsigned int total_valid_inode_count; /* valid inode count */
int active_logs; /* # of active logs */
int dir_level; /* directory level */
+ int serialized_dio_pages; /* serialized direct IO pages */
block_t user_block_count; /* # of user blocks */
block_t total_valid_block_count; /* # of valid blocks */
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index a2e3a8f..4a2e51e 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -218,6 +218,7 @@ F2FS_RW_ATTR(NM_INFO, f2fs_nm_info, ram_thresh, ram_thresh);
F2FS_RW_ATTR(NM_INFO, f2fs_nm_info, ra_nid_pages, ra_nid_pages);
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, serialized_dio_pages, serialized_dio_pages);
F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, cp_interval, cp_interval);
#define ATTR_LIST(name) (&f2fs_attr_##name.attr)
@@ -234,6 +235,7 @@ static struct attribute *f2fs_attrs[] = {
ATTR_LIST(min_fsync_blocks),
ATTR_LIST(max_victim_search),
ATTR_LIST(dir_level),
+ ATTR_LIST(serialized_dio_pages),
ATTR_LIST(ram_thresh),
ATTR_LIST(ra_nid_pages),
ATTR_LIST(cp_interval),
@@ -1125,6 +1127,7 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
atomic_set(&sbi->nr_pages[i], 0);
sbi->dir_level = DEF_DIR_LEVEL;
+ sbi->serialized_dio_pages = DEF_SERIALIZED_DIO_PAGES;
sbi->cp_interval = DEF_CP_INTERVAL;
clear_sbi_flag(sbi, SBI_NEED_FSCK);
--
2.6.3
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/