I am referring to this post.
Tuning nginx worker_process to obtain 100k hits per min
I have 3 8 core servers for a total of 24 cores. Currently doing 6K qps. In my nginx config I have one worker per core. For uWSGI I am using the gevent loop and I have one process per core. I set nice for nginx to -2 and nice for uWSGI to -1.
So..what is the better explanation for setting the number of workers to two times the number of cpu's? I was always with the school of thought of one proccess per core. I would love to increase if it will help. Maybee number of cores plus half? i.e 12? I would love to get more juice out of my machines. Latency is very critical. I am a 100ms deadline to respond to a post. So far working great. Think time is 5ms but still....how can i better tune...
user www-data;
worker_processes 8;
worker_priority -2;
worker_rlimit_nofile 200000;
events {
worker_connections 30000;
multi_accept on;
use epoll;
}