Hi,
I am trying to get some metrics on my installation. For that I have
disabled the database (db=None) and I am requesting a very simple
controller:
def hello1():
return "Hello World"
I have a multi-threaded test program which is sending requests to web2py
(using WebClient). I have let my test run for several iterations, and I got
the following metrics:
CLIENTS ROUNDS ELLAPSED (s) TOT REQ AVG (ms) REQ/s
8 72 17.969 576 31.197 32.05
9 25 6.973 225 30.991 32.27
6 4 0.748 24 31.179 32.07
7 17 3.645 119 30.634 32.64
7 58 12.340 406 30.393 32.90
3 73 6.751 219 30.827 32.44
10 12 3.399 120 28.329 35.30
4 32 3.941 128 30.791 32.48
1 27 0.742 27 27.496 36.37
8 58 14.156 464 30.508 32.78
4 31 3.574 124 28.819 34.70
8 14 3.274 112 29.232 34.21
9 68 18.722 612 30.591 32.69
9 81 22.070 729 30.275 33.03
4 47 6.031 188 32.079 31.17
7 75 15.491 525 29.506 33.89
1 90 2.495 90 27.719 36.08
8 38 9.495 304 31.234 32.02
2 60 4.000 120 33.335 30.00
2 81 5.057 162 31.214 32.04
(CLIENTS is the number of active threads, ROUNDS is the number of requests
that each thread performs)
As you see, I am getting quite stable speed of between 30 and 36 requests/s.
This is fine, but 32 req/s is quite low. My system is (this is my
development system, an HP 635):
- web2py 2.0.2
- 2 cores, AMD E-450
- 4 GB RAM
- Linux 2.6.32-5-686
- CrunchBang Linux statler
- Python 2.7.2
- I am using Rocket
- No apache (or any other frontend). Directly accessing localhost:8000
- web2py and my test program are sharing the same machine.
I would like to optimize these parameters before moving to production. Is
there something obvious that I could try to improve these metrics?
Is 30 req/s an expected value for such a simple test? I expected more than
100 req/s.
Now that I am writing and thinking about this, I am not even sure if the
limiting factor is web2py or my test program ... How could I know which
party is "saturating"?
Thanks,
Daniel
--
