Performance drops a lot in a windows environment

[yuhangch]

Hi, I originally wanted to discuss it here, but I didn't find the entrance. Thank you for providing this link.

I made a simple application based on rio-viz, to provide
services for multiple local cog data, I create new COGReader in each request as following：

async def _tiles(ctx: TilesContext) :
    path = root + ctx.paths[0]
    with COGReader(path) as dataset:

        tile, mask = dataset.tile(
            tile_x=ctx.x,
            tile_y=ctx.y,
            tile_z=ctx.z,
            indexes=ctx.indexes
        )

    content = await _render(tile, mask, ctx)
    return content

@routes.get('/{path}/tiles/{z}/{x}/{y}.{ext}')
async def tiles(request: Request):
    start = datetime.datetime.now()
    ctx = TileContext(request)

    # ···
    content = await _tiles(ctx)
    # ···
    end = datetime.datetime.now()
    print(type(content), start, end, (end - start).microseconds / 1000)
    return TileResponse(content, ext)

I deployed it on our windows platform server, found that the performance drops a lot in a windows environment.

With wsl(ubuntu 18.04) env:

• rasterio==1.1.8 (install via pypi)
• rio-tiler==2.0.0rc3

# packages in environment at /home/git/miniconda3/envs/rio118:
#
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main
affine                    2.3.0                    pypi_0    pypi
aiohttp                   3.7.3                    pypi_0    pypi
aiohttp-cors              0.7.0                    pypi_0    pypi
async-timeout             3.0.1                    pypi_0    pypi
attrs                     20.3.0                   pypi_0    pypi
boto3                     1.16.40                  pypi_0    pypi
botocore                  1.19.40                  pypi_0    pypi
ca-certificates           2020.12.8            h06a4308_0
certifi                   2020.12.5        py38h06a4308_0
chardet                   3.0.4                    pypi_0    pypi
click                     7.1.2                    pypi_0    pypi
click-plugins             1.1.1                    pypi_0    pypi
cligj                     0.7.1                    pypi_0    pypi
idna                      2.10                     pypi_0    pypi
jmespath                  0.10.0                   pypi_0    pypi
ld_impl_linux-64          2.33.1               h53a641e_7
libedit                   3.1.20191231         h14c3975_1
libffi                    3.3                  he6710b0_2
libgcc-ng                 9.1.0                hdf63c60_0
libstdcxx-ng              9.1.0                hdf63c60_0
mercantile                1.1.6                    pypi_0    pypi
morecantile               2.0.1                    pypi_0    pypi
multidict                 5.1.0                    pypi_0    pypi
ncurses                   6.2                  he6710b0_1
numexpr                   2.7.1                    pypi_0    pypi
numpy                     1.19.4                   pypi_0    pypi
openssl                   1.1.1i               h27cfd23_0
pip                       20.3.3           py38h06a4308_0
pydantic                  1.7.3                    pypi_0    pypi
pyparsing                 2.4.7                    pypi_0    pypi
python                    3.8.5                h7579374_1
python-dateutil           2.8.1                    pypi_0    pypi
python-dotenv             0.15.0                   pypi_0    pypi
rasterio                  1.1.8                    pypi_0    pypi
readline                  8.0                  h7b6447c_0
redis                     3.5.3                    pypi_0    pypi
requests                  2.25.1                   pypi_0    pypi
rio-color                 1.0.0                    pypi_0    pypi
rio-mucho                 1.0.0                    pypi_0    pypi
rio-tiler                 2.0.0rc3                 pypi_0    pypi
s3transfer                0.3.3                    pypi_0    pypi
setuptools                51.0.0           py38h06a4308_2
six                       1.15.0                   pypi_0    pypi
snuggs                    1.4.7                    pypi_0    pypi
sqlite                    3.33.0               h62c20be_0
tk                        8.6.10               hbc83047_0
typing-extensions         3.7.4.3                  pypi_0    pypi
urllib3                   1.26.2                   pypi_0    pypi
wheel                     0.36.2             pyhd3eb1b0_0
xz                        5.2.5                h7b6447c_0
yarl                      1.6.3                    pypi_0    pypi
zlib                      1.2.11               h7b6447c_3

processing time for each tile is about +-100 millisecond

(rio118) git@Data_share:/mnt/e/viz-server/viz$ python viz.py --port 8084
8084
init done
======= Serving on http://0.0.0.0:8084/ ======
/home/git/miniconda3/envs/rio118/lib/python3.8/site-packages/rasterio/io.py:133: NotGeoreferencedWarning: Dataset has no geotransform set. The identity matrix may be returned.
  return writer(mempath, 'w+', driver=driver, width=width,
/dom_cog.tif/tiles/22/3445633/1651205.png  processing
<class 'bytes'> 2020-12-19 11:03:57.299616 2020-12-19 11:03:57.584436 284.82
/dom_cog.tif/tiles/22/3445633/1651206.png  processing
<class 'bytes'> 2020-12-19 11:03:57.585221 2020-12-19 11:03:57.676697 91.476
/dom_cog.tif/tiles/21/1722815/825603.png  processing
<class 'bytes'> 2020-12-19 11:03:57.678025 2020-12-19 11:03:57.741681 63.656
/dom_cog.tif/tiles/21/1722814/825601.png  processing
<class 'bytes'> 2020-12-19 11:03:57.742787 2020-12-19 11:03:57.819225 76.438
/dom_cog.tif/tiles/22/3445629/1651202.png  processing
<class 'bytes'> 2020-12-19 11:03:57.820257 2020-12-19 11:03:57.943710 123.453
/dom_cog.tif/tiles/22/3445628/1651202.png  processing
<class 'bytes'> 2020-12-19 11:03:57.944350 2020-12-19 11:03:58.055123 110.773
/dom_cog.tif/tiles/21/1722814/825603.png  processing
<class 'bytes'> 2020-12-19 11:03:58.055948 2020-12-19 11:03:58.141774 85.826
/dom_cog.tif/tiles/21/1722815/825602.png  processing
<class 'bytes'> 2020-12-19 11:03:58.142937 2020-12-19 11:03:58.235090 92.153
/dom_cog.tif/tiles/21/1722815/825601.png  processing
<class 'bytes'> 2020-12-19 11:03:58.236292 2020-12-19 11:03:58.324430 88.138

With windows env:

• rasterio==1.1.8 (install via conda-forge)
• rio-tiler==2.0.0rc3

(base) E:\viz-server\viz>conda list
# packages in environment at C:\Users\dell\miniconda3:
#
# Name                    Version                   Build  Channel
affine                    2.3.0                      py_0    conda-forge
aiohttp                   3.7.3                    pypi_0    pypi
aiohttp-cors              0.7.0                    pypi_0    pypi
async-timeout             3.0.1                    pypi_0    pypi
attrs                     20.3.0             pyhd3deb0d_0    conda-forge
boost-cpp                 1.74.0               h54f0996_1    conda-forge
boto3                     1.16.37                  pypi_0    pypi
botocore                  1.19.37                  pypi_0    pypi
bzip2                     1.0.8                h8ffe710_4    conda-forge
ca-certificates           2020.12.5            h5b45459_0    conda-forge
cairo                     1.16.0            hd28d34b_1006    conda-forge
certifi                   2020.12.5        py38haa244fe_0    conda-forge
cffi                      1.14.0           py38h7a1dbc1_0
cfitsio                   3.470                h0af3d06_7    conda-forge
chardet                   3.0.4                 py38_1003
click                     7.1.2              pyh9f0ad1d_0    conda-forge
click-plugins             1.1.1                      py_0    conda-forge
cligj                     0.7.1              pyhd8ed1ab_0    conda-forge
conda                     4.9.2            py38haa244fe_0    conda-forge
conda-package-handling    1.6.1            py38h62dcd97_0
console_shortcut          0.1.1                         4
cryptography              2.9.2            py38h7a1dbc1_0
curl                      7.71.1               h4b64cdc_8    conda-forge
expat                     2.2.9                h33f27b4_2    conda-forge
freetype                  2.10.4               h546665d_0    conda-forge
freexl                    1.0.5             hd288d7e_1002    conda-forge
geos                      3.8.1                he025d50_0    conda-forge
geotiff                   1.6.0                h8884d1a_3    conda-forge
gettext                   0.19.8.1          h1a89ca6_1005    conda-forge
glib                      2.66.3               h0e60522_1    conda-forge
hdf4                      4.2.13            h0e5069d_1004    conda-forge
hdf5                      1.10.6          nompi_h5268f04_1112    conda-forge
icu                       67.1                 h33f27b4_0    conda-forge
idna                      2.9                        py_1
intel-openmp              2020.3             h57928b3_311    conda-forge
jmespath                  0.10.0                   pypi_0    pypi
jpeg                      9d                   h8ffe710_0    conda-forge
kealib                    1.4.14               ha3510f1_0    conda-forge
krb5                      1.17.2               hbae68bd_0    conda-forge
libblas                   3.8.0                    21_mkl    conda-forge
libcblas                  3.8.0                    21_mkl    conda-forge
libcurl                   7.71.1               h4b64cdc_8    conda-forge
libffi                    3.3                  h0e60522_2    conda-forge
libgdal                   3.1.4                h0e5aa5a_0    conda-forge
libglib                   2.66.3               h8c8e623_1    conda-forge
libiconv                  1.16                 he774522_0    conda-forge
libkml                    1.3.0             he9e54da_1012    conda-forge
liblapack                 3.8.0                    21_mkl    conda-forge
libnetcdf                 4.7.4           nompi_h3a9aa94_107    conda-forge
libpng                    1.6.37               h1d00b33_2    conda-forge
libpq                     12.3                 h4f54205_3    conda-forge
libspatialite             5.0.0                hf693123_0    conda-forge
libssh2                   1.9.0                hb06d900_5    conda-forge
libtiff                   4.1.0                hc10be44_6    conda-forge
libwebp-base              1.1.0                h8ffe710_3    conda-forge
libxml2                   2.9.10               h1006b36_2    conda-forge
lz4-c                     1.9.2                h62dcd97_2    conda-forge
menuinst                  1.4.16           py38he774522_0
mercantile                1.1.6                    pypi_0    pypi
mkl                       2020.4             hb70f87d_311    conda-forge
morecantile               2.0.1                    pypi_0    pypi
multidict                 5.1.0                    pypi_0    pypi
numexpr                   2.7.1                    pypi_0    pypi
numpy                     1.19.4           py38h0cc643e_1    conda-forge
openjpeg                  2.3.1                h48faf41_3    conda-forge
openssl                   1.1.1i               h8ffe710_0    conda-forge
pcre                      8.44                 ha925a31_0    conda-forge
pip                       20.0.2                   py38_3
pixman                    0.40.0               h8ffe710_0    conda-forge
poppler                   0.89.0               h5d62644_1    conda-forge
poppler-data              0.4.10                        0    conda-forge
postgresql                12.3                 h0f1a9bc_3    conda-forge
powershell_shortcut       0.0.1                         3
proj                      7.1.1                h7d85306_3    conda-forge
pycosat                   0.6.3            py38he774522_0
pycparser                 2.20                       py_0
pydantic                  1.7.3                    pypi_0    pypi
pyopenssl                 19.1.0                   py38_0
pyparsing                 2.4.7              pyh9f0ad1d_0    conda-forge
pysocks                   1.7.1                    py38_0
python                    3.8.3                he1778fa_0
python-dateutil           2.8.1                    pypi_0    pypi
python-dotenv             0.15.0                   pypi_0    pypi
python_abi                3.8                      1_cp38    conda-forge
pywin32                   227              py38he774522_1
rasterio                  1.1.8            py38h5653988_0    conda-forge
redis                     3.5.3                    pypi_0    pypi
requests                  2.23.0                   py38_0
rio-color                 1.0.0                    pypi_0    pypi
rio-mucho                 1.0.0                    pypi_0    pypi
rio-tiler                 2.0.0rc3                 pypi_0    pypi
ruamel_yaml               0.15.87          py38he774522_0
s3transfer                0.3.3                    pypi_0    pypi
setuptools                46.4.0                   py38_0
six                       1.14.0                   py38_0
snuggs                    1.4.7                      py_0    conda-forge
sqlite                    3.34.0               h8ffe710_0    conda-forge
tiledb                    2.1.3                h968eb34_0    conda-forge
tk                        8.6.10               h8ffe710_1    conda-forge
tqdm                      4.46.0                     py_0
typing-extensions         3.7.4.3                  pypi_0    pypi
urllib3                   1.25.8                   py38_0
vc                        14.1                 h0510ff6_4
vs2015_runtime            14.16.27012          hf0eaf9b_1
wheel                     0.34.2                   py38_0
win_inet_pton             1.1.0                    py38_0
wincertstore              0.2                      py38_0
xerces-c                  3.2.3                ha925a31_1    conda-forge
xz                        5.2.5                h62dcd97_1    conda-forge
yaml                      0.1.7                hc54c509_2
yarl                      1.6.3                    pypi_0    pypi
zlib                      1.2.11               h62dcd97_4
zstd                      1.4.5                h1f3a1b7_2    conda-forge

processing time (same image) for each tile is about +-500 millisecond

(base) E:\viz-server\viz>python viz.py --port 8084
======= Serving on http://0.0.0.0:8084/ ======
C:\Users\dell\miniconda3\lib\site-packages\rasterio\io.py:133: NotGeoreferencedWarning: Dataset has no geotransform set. The identity matrix may be returned.
  return writer(mempath, 'w+', driver=driver, width=width,
/dom_cog.tif/tiles/22/3445623/1651205.png  processing
<class 'bytes'> 2020-12-19 11:00:43.371836 2020-12-19 11:00:48.840684 468.848
/dom_cog.tif/tiles/22/3445623/1651204.png  processing
<class 'bytes'> 2020-12-19 11:00:48.840684 2020-12-19 11:00:51.387426 546.742
/dom_cog.tif/tiles/21/1722813/825601.png  processing
<class 'bytes'> 2020-12-19 11:00:51.387426 2020-12-19 11:00:54.949573 562.147
/dom_cog.tif/tiles/22/3445627/1651208.png  processing
<class 'bytes'> 2020-12-19 11:00:54.949573 2020-12-19 11:00:57.449617 500.044
/dom_cog.tif/tiles/22/3445628/1651208.png  processing
<class 'bytes'> 2020-12-19 11:00:57.449617 2020-12-19 11:00:59.965225 515.608
/dom_cog.tif/tiles/21/1722813/825602.png  processing
<class 'bytes'> 2020-12-19 11:00:59.965225 2020-12-19 11:01:03.496726 531.501
/dom_cog.tif/tiles/21/1722813/825603.png  processing
<class 'bytes'> 2020-12-19 11:01:03.496726 2020-12-19 11:01:07.027833 531.107
/dom_cog.tif/tiles/21/1722812/825602.png  processing
<class 'bytes'> 2020-12-19 11:01:07.027833 2020-12-19 11:01:10.543423 515.59
/dom_cog.tif/tiles/20/861406/412801.png  processing
<class 'bytes'> 2020-12-19 11:01:10.543423 2020-12-19 11:01:13.889678 346.255

My question is, Is there some suggestions to improve the performance in the windows environment?
We prefer it to be deployed in the windows environment because it is easy to manage, or we have to deploy it in the ubuntu environment.
Thanks.

[vincentsarago]

@yuhangch thanks for the question.

I have 0 knowledge in windows, so I'm not sure I have an answer. If I were you I will start first by profiling the function https://docs.python.org/3.8/library/profile.html and then also make sure the GDAL environment variable a set to maximize performances! (e.g https://github.com/developmentseed/titiler/blob/master/deployment/AWS/cdk/app.py#L21-L26)

[yuhangch]

Thanks🙏 ， I'll try the steps above.

[vincentsarago]

@yuhangch just found https://github.com/sunhailin-Leo/fastapi_profiler which seems a nice module to profile rio-viz like API (based on fastapi).