Flask API server slow response time - gunicorn

Question

I created an API server with Flask, I use gunicorn with eventlet to run it. I noticed a long response time from Flask server when calling APIs. I did a profiling with my client, one ran from my laptop, one ran directly in Flask API server.
From my laptop:
302556 function calls (295712 primitive calls) in 5.594 seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
72 4.370 0.061 4.370 0.061 {method 'poll' of 'select.epoll' objects}
16 0.374 0.023 0.374 0.023 {method 'connect' of '_socket.socket' objects}
16 0.213 0.013 0.213 0.013 {method 'load_verify_locations' of '_ssl._SSLContext' objects}
16 0.053 0.003 0.058 0.004 httplib.py:798(close)
52 0.034 0.001 0.034 0.001 {method 'do_handshake' of '_ssl._SSLSocket' objects}
On server:
231449 function calls (225936 primitive calls) in 3.320 seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
12 2.132 0.178 2.132 0.178 {built-in method read}
13 0.286 0.022 0.286 0.022 {method 'poll' of 'select.epoll' objects}
12 0.119 0.010 0.119 0.010 {_ssl.sslwrap}
12 0.095 0.008 0.095 0.008 {built-in method do_handshake}
855/222 0.043 0.000 0.116 0.001 sre_parse.py:379(_parse)
1758/218 0.029 0.000 0.090 0.000 sre_compile.py:32(_compile)
1013 0.027 0.000 0.041 0.000 sre_compile.py:207(_optimize_charset)
12429 0.023 0.000 0.029 0.000 sre_parse.py:182(__next)
So, I saw my client took long time to wait from server response base on the profile result.
I used gunicorn with eventlet to serve Flask app, with the flowing configuration:
import multiprocessing
bind = ['0.0.0.0:8000']
backlog = 2048
workers = multiprocessing.cpu_count() * 2 + 1
worker_class = 'eventlet'
user = 'www-data'
group = 'www-data'
loglevel = 'info'
My client is an custom HTTP client using eventlet to patch httplib2 and create a pool to connect to server.
I stuck here with the troubleshooting. All server stats were normal. How can I detect the bottle neck of my API server?