How to tell: Is it nginx or PHP-cgi which is slower?

Question

I run a website with a lot of traffic. I'm using nginx and php fast-cgi. However, when loading simple php documents in the browser, it takes up to 5-6 seconds before the page is loaded. Sometimes it even just results in a "404 not found" nginx error.

When using "htop" on debian, I am nowhere near using the ressources of my server. So what's going on? Is it nginx? Is it PHP-fast-cgi? I can't tell where the bottle neck is.

A snippet of my nginx config:

user  www-data www-data;
worker_processes  2;

events {
    worker_connections  4048;
}


http {
    include       mime.types;
    default_type  application/octet-stream;
    sendfile        on;
    #tcp_nopush     on;
    keepalive_timeout  10;
    include /usr/local/nginx/sites-enabled/*;
...
}

A snippet of my fast-cgi config:

EXEC_AS_USER=www-data

# Host and TCP port for FASTCGI-Listener (default: localhost:9000)

FCGI_HOST=localhost
FCGI_PORT=9000

# Environment variables, which are processed by PHP

PHP_FCGI_CHILDREN=8
PHP_FCGI_MAX_REQUESTS=0

Thank you very much!!

Answer 1

How much is a lot traffic? Just a few tips:

• use cache as much as possible
• the number of worker processes could be equal to number of CPU cores
• close live connections as early as possible or disable them completely (try let's say keepalive_timeout 5)
• each connection needs a filehandle worker_rlimit_nofile 20000;
• total amount of users you can serve in 1 second (approximately) = worker_processes * worker_connections/ (keepalive_timeout * 2)
• disable access logs unless you really need them
• use a balancer (if it's not feasible with one server)

Answer 2

Quiet the server, put strace on all the php and nginx processes, try a request and watch where it waits.

Alternatively, look for a profiling tool for php and gather some data with it.

Answer 3

1. Try to configure FastCGI as unix socket and not internet socket. This would save you in some cases up to 20% of time spent.
2. You can push your worker_processes higher up to number of available cores.
3. Check your php-fpm config as it has enough options to:
   • kill very long running requests;
   • report about long running request.
4. Change PHP_FCGI_MAX_REQUESTS to something relevant like 1000. Just for the case of memory corruption.