Re: [PATCH] scripts/simplebench: compare write request performance

[Andrey Shinkevich]

On 24.06.2020 08:45, Andrey Shinkevich wrote:

> The script 'bench_write_req.py' allows comparing performances of write
> request for two qemu-img binary files.
>
> Suggested-by: Denis V. Lunev <den@openvz.org>
> Suggested-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
> ---
>  scripts/simplebench/bench_write_req.py | 150 +++++++++++++++++++++++++++++++++
>  1 file changed, 150 insertions(+)
>  create mode 100755 scripts/simplebench/bench_write_req.py
>
> diff --git a/scripts/simplebench/bench_write_req.py b/scripts/simplebench/bench_write_req.py
> new file mode 100755
> index 0000000..e127dcc
> --- /dev/null
> +++ b/scripts/simplebench/bench_write_req.py
> @@ -0,0 +1,150 @@
> +#!/usr/bin/env python3
>     
    ...
> +    # Test-cases are "rows" in benchmark resulting table, 'id' is a caption
> +    # for the row, other fields are handled by bench_func.
> +    test_cases = [
> +        {
> +            'id': '<cluster middle>',
> +            'block_size': 4096,
> +            'block_offset': 524288,
> +            'requests': 100
>     
    s/100/10/
> +        },
> +        {
> +            'id': '<cluster overlap>',
> +            'block_size': 524288,
> +            'block_offset': 4096,
> +            'requests': 10
>     
    s/10/2/
> +        },
> +    ]
>

I reduced the number of requests per the test case to make the script runtime more reasonable for HDD device and got the following results:

qemu-img1 is the current version of QEMU

qemu-img2 is the current version but without the feature "[PATCH v14 1/1] qcow2: skip writing zero buffers to empty COW areas" (git commit ID: c8bb23cbdbe32f5)

SSD:

$ time ./scripts/simplebench/bench_write_req.py ./qemu-img1 ./qemu-img2 image.qcow2
Testing 1/4: <qemu-img binary 1> :: <cluster middle>
  #run 1
    {'seconds': 0.377}
  #run 2
    {'seconds': 0.382}
  #run 3
    {'seconds': 0.376}
Testing 2/4: <qemu-img binary 1> :: <cluster overlap>
  #run 1
    {'seconds': 6.836}
  #run 2
    {'seconds': 8.513}
  #run 3
    {'seconds': 6.843}
Testing 3/4: <qemu-img binary 2> :: <cluster middle>
  #run 1
    {'seconds': 11.46}
  #run 2
    {'seconds': 8.734}
  #run 3
    {'seconds': 8.579}
Testing 4/4: <qemu-img binary 2> :: <cluster overlap>
  #run 1
    {'seconds': 11.848}
  #run 2
    {'seconds': 17.8}
  #run 3
    {'seconds': 19.822}
Done
-----------------  -------------------  -------------------
                   <qemu-img binary 1>  <qemu-img binary 2>
<cluster middle>   0.38 +- 0.00         9.59 +- 1.87
<cluster overlap>  7.40 +- 1.12         16.49 +- 4.64
-----------------  -------------------  -------------------

real    1m43.769s
user    0m1.881s
sys 0m4.677s

HDD:

$ sudo time ./scripts/simplebench/bench_write_req.py ./qemu-img1 ./qemu-img2 /vz/tmp/image.qcow2
Testing 1/4: <qemu-img binary 1> :: <cluster middle>
  #run 1
    {'seconds': 54.738}
  #run 2
    {'seconds': 57.049}
  #run 3
    {'seconds': 58.199}
Testing 2/4: <qemu-img binary 1> :: <cluster overlap>
  #run 1
    {'seconds': 12.846}
  #run 2
    {'seconds': 11.015}
  #run 3
    {'seconds': 11.11}
Testing 3/4: <qemu-img binary 2> :: <cluster middle>
  #run 1
    {'seconds': 53.969}
  #run 2
    {'seconds': 58.22}
  #run 3
    {'seconds': 61.413}
Testing 4/4: <qemu-img binary 2> :: <cluster overlap>
  #run 1
    {'seconds': 15.702}
  #run 2
    {'seconds': 12.627}
  #run 3
    {'seconds': 11.879}
Done
-----------------  -------------------  -------------------
                   <qemu-img binary 1>  <qemu-img binary 2>
<cluster middle>   56.66 +- 1.92        57.87 +- 3.90
<cluster overlap>  11.66 +- 1.19        13.40 +- 2.30
-----------------  -------------------  -------------------
3.31user 7.93system 7:03.13elapsed 2%CPU (0avgtext+0avgdata 78280maxresident)k
50240inputs+22659372outputs (0major+965543minor)pagefaults 0swaps

Andrey