Re: [PATCH for-5.16 v3 0/8] task work optimization

[Pavel Begunkov <asml.silence@gmail.com>]

On 10/27/21 15:02, Hao Xu wrote:
> Tested this patchset by manually replace __io_queue_sqe() in
> io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works.
> Then test with fio:

If submissions and completions are done by the same task it doesn't
really matter in which order they're executed because the task won't
get back to userspace execution to see CQEs until tw returns.
Furthermore, it even might be worse because the earlier you submit
the better with everything else equal.

IIRC, that's how it's with fio, right? If so, you may get better
numbers with a test that does submissions and completions in
different threads.

Also interesting to find an explanation for you numbers assuming
they're stable. 7/8 batching? How often it does it go this path?
If only one task submits requests it should already be covered
with existing batching.


> ioengine=io_uring
> sqpoll=1
> thread=1
> bs=4k
> direct=1
> rw=randread
> time_based=1
> runtime=600
> randrepeat=0
> group_reporting=1
> filename=/dev/nvme0n1
> 
> 2/8 set unlimited priority_task_list, 8/8 set a limitation of
> 1/3 * (len_prority_list + len_normal_list), data below:
>     depth     no 8/8   include 8/8      before this patchset
>      1        7.05         7.82              7.10
>      2        8.47         8.48              8.60
>      4        10.42        9.99              10.42
>      8        13.78        13.13             13.22
>      16       27.41        27.92             24.33
>      32       49.40        46.16             53.08
>      64       102.53       105.68            103.36
>      128      196.98       202.76            205.61
>      256      372.99       375.61            414.88
>      512      747.23       763.95            791.30
>      1024     1472.59      1527.46           1538.72
>      2048     3153.49      3129.22           3329.01
>      4096     6387.86      5899.74           6682.54
>      8192     12150.25     12433.59          12774.14
>      16384    23085.58     24342.84          26044.71
> 
> It seems 2/8 is better, haven't tried other choices other than 1/3,
> still put 8/8 here for people's further thoughts.
> 
> Hao Xu (8):
>    io-wq: add helper to merge two wq_lists
>    io_uring: add a priority tw list for irq completion work
>    io_uring: add helper for task work execution code
>    io_uring: split io_req_complete_post() and add a helper
>    io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
>    io_uring: add nr_ctx to record the number of ctx in a task
>    io_uring: batch completion in prior_task_list
>    io_uring: add limited number of TWs to priority task list
> 
>   fs/io-wq.h    |  21 +++++++
>   fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------
>   2 files changed, 138 insertions(+), 51 deletions(-)
> 

-- 
Pavel Begunkov