The following stats are from a site where nginx was placed in front of a webserver to deliver static files and let the CMS application on the legacy webserver focus on the dynamic work. This is done via the nginx HttpProxyModule on the same physical machine.
The stats are created by the nginx_combined_v2 munin plugin.
The main HTML file loads about 8 CSS files, about 10 js files (and about 5 more, but from external sites) and almost 200 inline elements which results in up to 2-4 megs of data. Don't ask.
There's a lot of caching involved to handle the load and the user experience is fine (for the non-mobile users, of course ;-)
I wonder if I should worry about the high values of "Waiting" connections or if this is just a usual pattern when such huge numbers of inline-elements are involved?
As the legacy server has less load now than before I don't assume that nginx has to wait a long time for the backend.
The stats by week show nicely the peeks during the day when people should actually work not surf the net ;-)
Waiting connections is not a very big problem - it's connections in keepalive state. But they are still using some system resources (memory for storing file handle, network buffers, etc). So it's best to keep keepalive_timeout as low as possible (but lower values might require some users to re-establish connections often, so you'll need to find out good value for your workload).