> -----Original Message-----
> From: changfeng...@vivo.com <changfeng...@vivo.com> On Behalf Of
> Chao Yu
> Sent: Monday, November 8, 2021 10:21 PM
> To: 常凤楠 <changfeng...@vivo.com>; jaeg...@kernel.org
> Cc: linux-f2fs-devel@lists.sourceforge.net
> Subject: Re: Do we need serial io for compress file?
>
> On 2021/11/8 11:54, Fengnan Chang wrote:
> > In my test, serial io for compress file will make multithread small
> > write performance drop a lot.
> >
> > I'm try to fingure out why we need __should_serialize_io, IMO, we use
> > __should_serialize_io to avoid deadlock or try to improve sequential
> > performance, but I don't understand why we should do this for
>
> It was introduced to avoid fragmentation of file blocks.
So, for small write on compress file, is this still necessary? I think we
should treat compress file as regular file.
>
> > compressed file. In my test, if we just remove this, write same file
> > in multithread will have problem, but parallel write different files
> > in multithread
>
> What do you mean by "write same file in multithread will have problem"?
If just remove compress file in __should_serialize_io()
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index f4fd6c246c9a..7bd429b46429 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -3165,8 +3165,8 @@ static inline bool __should_serialize_io(struct inode
*inode,
if (IS_NOQUOTA(inode))
return false;
- if (f2fs_need_compress_data(inode))
- return true;
+ //if (f2fs_need_compress_data(inode))
+ // return true;
if (wbc->sync_mode != WB_SYNC_ALL)
return true;
if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
and use fio to start multi thread to write same file, fio will hung.
fio.conf:
[global]
direct=1
numjobs=8
time_based
runtime=30
ioengine=sync
iodepth=16
buffer_pattern="ZZZZ"
fsync=1
[file0]
name=fio-rand-RW
filename=fio-rand-RW
rw=rw
rwmixread=60
rwmixwrite=40
bs=1M
size=64M
[file1]
name=fio-rand-RW
filename=fio-rand-RW
rw=randrw
rwmixread=60
rwmixwrite=40
bs=4K
size=64M
>
> Thanks,
>
> > is ok. So I think maybe we should use another lock to allow write
> different files in multithread.
> >
_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
