Here is the secret to getting the whole nginx/fastcgi vs. apache/mod_php thing to click in your head.
The trick is to think of it in two layers. The "spoonfeeding" layer and the "worker" layer.
In nginx/fastcgi, nginx is the spoonfeeding layer and the fastcgi worker procceses are, you guessed it, the workers. Nginx effectively acts as a proxy, holding open the thousands of connections users have made but only a small fraction of them are actually being worked on by the worker processes at any given moment.
With apache/mod_php there is no spoonfeeding layer. Every single apache child has the php interpreter loaded and acts as a worker process. When you look at your nginx vs. your apache setups think of apache children as the fascgi workers not as nginx.
Most of the time people switch to nginx and claim amazing results what got them that was the addition of a spoonfeeding layer. Technically, you could have done that several other ways. Most higher end sites who've had to care about this stuff for years let the loadbalancer handle the spoonfeeder layer, that way the (web)appservers can focus on being appservers. Doesn't matter if its a $120K Netscaler or a $0 copy of varnish or ha-proxy bound to localhost, the point is you never want a 1:1 relationship between "active but idle user connections" and "worker processes".
The trick is to think of it in two layers. The "spoonfeeding" layer and the "worker" layer.
In nginx/fastcgi, nginx is the spoonfeeding layer and the fastcgi worker procceses are, you guessed it, the workers. Nginx effectively acts as a proxy, holding open the thousands of connections users have made but only a small fraction of them are actually being worked on by the worker processes at any given moment.
With apache/mod_php there is no spoonfeeding layer. Every single apache child has the php interpreter loaded and acts as a worker process. When you look at your nginx vs. your apache setups think of apache children as the fascgi workers not as nginx.
Most of the time people switch to nginx and claim amazing results what got them that was the addition of a spoonfeeding layer. Technically, you could have done that several other ways. Most higher end sites who've had to care about this stuff for years let the loadbalancer handle the spoonfeeder layer, that way the (web)appservers can focus on being appservers. Doesn't matter if its a $120K Netscaler or a $0 copy of varnish or ha-proxy bound to localhost, the point is you never want a 1:1 relationship between "active but idle user connections" and "worker processes".