ping again.
>From my test, the patchset can improve the performance significantly.
The following data are tested on INTEL P4510 NVMe using fio with iodepth
128 and blocksize 4k.
----------------------------------------------------------------
| RANDREAD | RANDWRITE |
----------------------------------------------------------------
w/ async IO routines | 377MB/s | 405MB/s |
----------------------------------------------------------------
w/o async IO routines | 32.0MB/s | 62.3MB/s |
----------------------------------------------------------------
Regards,
Jiufei
On 2019/11/26 上午10:00, Jiufei Xue wrote:
> Hi miklos,
>
> Could you please kindly review this patch and give some advice?
>
> Thanks,
> Jiufei
>
> On 2019/11/20 下午5:45, Jiufei Xue wrote:
>> ovl stacks regular file operations now. However it doesn't implement
>> async IO routines and will convert async IOs to sync IOs which is not
>> expected.
>>
>> This patchset implements overlayfs async IO routines.
>>
>> Jiufei Xue (2)
>> vfs: add vfs_iocb_iter_[read|write] helper functions
>> ovl: implement async IO routines
>>
>> fs/overlayfs/file.c | 116 +++++++++++++++++++++++++++++++++++++++++------
>> fs/overlayfs/overlayfs.h | 2
>> fs/overlayfs/super.c | 12 ++++
>> fs/read_write.c | 58 +++++++++++++++++++++++
>> include/linux/fs.h | 16 ++++++
>> 5 files changed, 188 insertions(+), 16 deletions(-)
>>
>
