Debugging a production server?

 
What can I do to optimize my debugging ability on a production server (where
no slowdown during normal operation is acceptable)? Will compiling the
kernel with debug symbols and DDB cause any performance degradation? Any
other options I should consider? The goal is to be able to do some simple
investigation myself and write fully detailed PRs.
 
Also, how clever is the core dumping code? Will it be able to handle swap
spread out over two discs (as long as total size is larger than my RAM)? If
not, what about a situation where there's three swap areas, but only one of
them is bigger than the RAM?
 
/Daniel Eriksson