Apache benchmark: always a few "super late" requests, why ?

[henrique matias]

Perhaps no matter if i change number of workers, or worker connections,
there's always some super late connections on my "ab" tests..

Am very new to benchmark, but the way am doing now is:

ab -n 8000 -c 1000 http://address_to_a_plain_text_file

ab -n 8000 -c 1000
http://address_to_a_rails_address_that_queries_the_database

Almost all tests i do ends up with very good results up to 90% of the
requests ( both, with plain text and with rails script ), but then
sometimes in this last 10% there's very slow requests ( 10x more than the
fastest )..

For instance : ab -n 8000 -c 1000 http://address_to_a_plain_text_file

Percentage of the requests served within a certain time (ms)
  50%    142
  66%    152
  75%    156
  80%    158
  90%    169
  95%    181
  98%   4180
  99%   5236
 100%   5485 (longest request)

The ones that goes to the database sometimes get very slow compared to
plain text files, so am guessing i should enlarge somehow the pipe betweens
rails and the db.

For instance : ab -n 8000 -c 1000 http://address_to_rails_touching_db_file

Percentage of the requests served within a certain time (ms)
  50%     10
  66%     10
  75%     10
  80%     11
  90%     12
  95%     13
  98%     16
  99%     25
 100%  13237 (longest request)


Even my plain text files are suffering, so i would guess that is a problem
in my worker_process / worker_connections and keepalive_timeout ?


And for the file that touchs the database i would guess i should be
tweaking some database configuration between my rails and mongodb...


Any advices?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.nginx.org/pipermail/nginx/attachments/20130505/aa528dd1/attachment.html>