tile data read very slowly

[kobeyk]

Excuse me, I have a problem,I converted a cog tif with rio-cogeo and put it locally, then read it with rio-tiler, and ran it with py code alone, and found that the reading process took a long time, >600ms, in fact, normal tif is very fast when read with gdal, I would like to know why this piece is slow. Here's how long it takes to run alone：

code：

import os
os.environ['PROJ_LIB'] = r"D:\Anaconda\Anaconda3\envs\gdal3.6\Lib\site-packages\pyproj\proj_dir\share\proj"
import time
from rio_tiler.io import Reader
from rio_tiler.mosaic import mosaic_reader
def reader(file, x, y, z, **kwargs):
    startTime = time.perf_counter()
    image = Reader(file)
    elapsedTime = time.perf_counter() - startTime
    print(f"The code took {elapsedTime} seconds to read.")
    startTime = time.perf_counter()
    buff = image.tile(x, y, z, **kwargs)
    elapsedTime = time.perf_counter() - startTime
    print(f"The code took {elapsedTime} seconds to render.")
    image.close()
    return buff
cogList = [
   r"D:\gdm\image\shanghai\GF1D_PMS_E121.0_N31.9_20230523_L5A_1257227139\GF1D_PMS_E121.0_N31.9_20230523_L5B_1257227139.cog.tif",
]

x = 1714
y = 834
z = 11
img, assets = mosaic_reader(cogList, reader, x,y,z)
buff = img.render(img_format="webp")
with open("./images/slow.png", "wb") as f:
        f.write(buff)

In addition, I also used java to write a server to call python code, and wrote a front-end test. I found that cog as a image tile service, its performance is really a little too slow, I want to know, as cog data, whether it supports multi-user concurrent access? Is it possible to improve access efficiency by setting some parameters for tuning? In other words, in order to provide high-quality image tile services, it can only be carried out in accordance with the traditional slicing method. Below is a screenshot of the response time of the front-end call:

[kobeyk]

[vincentsarago]

@kobeyk can you try setting GDAL_DISABLE_READDIR_ON_OPEN=EMPTY_DIR in your env and see if this impact the performance?

ref: https://developmentseed.org/titiler/advanced/performance_tuning/

[kobeyk]

Thank you very much for your reply, I have tried this, but it did not improve the performance

[vincentsarago]

To be honest it's pretty hard for me to debug. I would need to know more about the data!

you can also checkout https://github.com/developmentseed/tilebench